SlideShare a Scribd company logo
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
DYNAMIC CLOSEST
COLOR WARPING TO
SORT AND COMPARE
PALETTES
© 2021 SIGGRAPH. ALL RIGHTS RESERVED.
SUZI KIM
SUNGHEE CHOI
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
WHAT IS A COLOR PALETTE?
A limited number of colors expressed in refined shapes
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 2
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
WHAT IS A COLOR PALETTE?
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 2
Intuitiveness
Generality
Easy-computation
Simplicity
A limited number of colors expressed in refined shapes
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
WHAT IS A COLOR PALETTE?
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 2
Intuitiveness
Generality
Easy-computation
Simplicity
A limited number of colors expressed in refined shapes
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
SORT THESE COLORS!
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 3
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
SORT THESE COLORS!
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 3
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
SORT THESE COLORS!
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 3
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
SORT THESE COLORS!
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 4
What if there are a lot of colors to be sorted?
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
SORT THESE COLORS!
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 4
What if there are a lot of colors to be sorted?
Or, what if you have to sort two palettes at the same time?
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 5
THIS PAPER INTRODUCES
considering both hue and lightness simultaneously
Single Palette Sorting
1
allowing palettes to be aligned concurrently
Multiple Palettes Sorting
2
reflecting sorted color tendencies
Similarity Measurement between Palettes
3
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Lexicographical Sorting according to the Color Space
Previous Sorting Methods #1
LEXICOGRAPHICAL SORTING
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 6
RGB
HSV
VHS
LAB
Pros
Cons  Between lightness and hue, only one priority can be employed.
 Hue is a circular quantity with discontinuity at 0 and 360 degrees.
 Simple and easy-to-compute
Input Palette
Sort
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Previous Sorting Methods #2
IMAGE COMPRESSION
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 7
Purpose Reflecting pixels’ spatiality to reproduce the original as much as possible while
increasing compression ratio.
Methods  Geometric graph traversal by reducing to the traveling salesman problem (TSP).
- Greedy algorithm, Neural network, Simulated annealing, Adaptive particle swarm
optimization, Genetic algorithm, etc.
Cons  Visual plausibility is not taken into account.
 It cannot sort multiple palettes simultaneously.
Color Sorting for the Image Compression
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Previous Sorting Methods #3
BINARY PALETTE SORTING
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 8
Cons
Purpose To sort two sets of multiple palettes
Methods  Similar to merge sort
 (1) Partitioning two given palettes into a single palette and (2) repeatedly sorting
merging in a bottom-up manner.
 It assumes that each palette has a similar color distribution pattern.
 Because the colors are sorted only in a row-chunk unit, it is difficult to arrange the
multiple palettes naturally if they are composed of unorganized colors.
Binary Palette Sorting (BPS) [Phan et al. 2018]
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
METHODS
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 9
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
SORTING STRATEGY
• We define a natural order of colors as adjacent colors have
smooth hue and lightness changes.
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 10
Image source: mbaumi @Unsplash
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
SORTING STRATEGY
• We define a natural order of colors as adjacent colors have
smooth hue and lightness changes.
• To reflect both conditions, we use a multi-dimensional graph
traversal with colors as vertices and connecting orders as edges.
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 10
Image source: cosminserban @Unsplash
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
SORTING STRATEGY
• We define a natural order of colors as adjacent colors have
smooth hue and lightness changes.
• To reflect both conditions, we use a multi-dimensional graph
traversal with colors as vertices and connecting orders as edges.
• We adopt CIELAB color space which is more perceptually
uniform than other color spaces.
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 10
Image source: Holger kkk Everding
CC BY-SA @Wikipedia
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
SORTING STRATEGY
• We define a natural order of colors as adjacent colors have
smooth hue and lightness changes.
• To reflect both conditions, we use a multi-dimensional graph
traversal with colors as vertices and connecting orders as edges.
• We adopt CIELAB color space which is more perceptually
uniform than other color spaces.
• Graph traversal is reduced to symmetric TSP using CIEDE2000
distance.
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 10
Image source: n-voitkevich @Pexels
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Method #1
SINGLE PALETTE SORTING
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 11
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Method #1
SINGLE PALETTE SORTING
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 11
TSP Solver
LKH
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Method #1
SINGLE PALETTE SORTING
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 12
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Method #2
SORTING A PAIR OF PALETTES
What if you need to sort two palettes at the same time?
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 13
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Method #2
SORTING A PAIR OF PALETTES
What if you need to sort two palettes at the same time?
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 13
We arrange the color swatches to satisfy two conditions:
Overall color flow is similar when the palettes are placed side by side.
Each palette is naturally sorted.
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Method #2
SORTING A PAIR OF PALETTES
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 14
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Method #2
SORTING A PAIR OF PALETTES
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 14
TSP Solver
LKH
Where to Cut?
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Method #2
SORTING A PAIR OF PALETTES
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 14
SPS Cutting
Strategy
TSP Solver
LKH
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Method #2
SORTING A PAIR OF PALETTES
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 15
SPS Cutting
Strategy
Input Palettes
Sorted Palettes
Merged Palettes
The most different pair of consecutive colors.
Radical color change cannot be avoided!
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Method #2
SORTING A PAIR OF PALETTES
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 15
Input Palettes
Cutting Strategy for a Pair of Palettes
 Arranging the two palettes in a similar color flow.
 Preventing sudden color changes in each palette.
Starting from the point of view of individual palettes.
Purpose
Methods
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Method #2
SORTING A PAIR OF PALETTES
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 15
Input Palettes
Cutting Strategy for a Pair of Palettes
 Arranging the two palettes in a similar color flow.
 Preventing sudden color changes in each palette.
Starting from the point of view of individual palettes.
Purpose
Methods
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Method #2
SORTING A PAIR OF PALETTES
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 15
Input Palettes
Cutting Strategy for a Pair of Palettes
 Arranging the two palettes in a similar color flow.
 Preventing sudden color changes in each palette.
Starting from the point of view of individual palettes.
Purpose
Methods
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Method #2
SORTING A PAIR OF PALETTES
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 16
Cutting Strategy for a Pair of Palettes
1. Find the longest edge connecting the same palette colors.
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Method #2
SORTING A PAIR OF PALETTES
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 16
Cutting Strategy for a Pair of Palettes
𝒄𝒂
𝒄𝒃
2. Among the two subpaths connecting 𝑐𝑎 and 𝑐𝑏, we choose the
one with fewer color vertices.
1. Find the longest edge connecting the same palette colors.
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Method #2
SORTING A PAIR OF PALETTES
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 16
Cutting Strategy for a Pair of Palettes
𝒄𝒂
𝒄𝒃
3. To decide which side the remaining colors should be bound
to, we find a point 𝑐𝑘 that minimizes
where 𝑎 ≤ 𝑘 ≤ 𝑏.
1. Find the longest edge connecting the same palette colors.
2. Among the two subpaths connecting 𝑐𝑎 and 𝑐𝑏, we choose the
one with fewer color vertices.
𝑖=𝑎+1
𝑘
𝑑 𝑐𝑎, 𝑐𝑖 +
𝑖=𝑘+1
𝑏−1
𝑑(𝑐𝑏, 𝑐𝑖) ,
𝒄𝒌
𝒄𝒌+𝟏
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Method #2
SORTING A PAIR OF PALETTES
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 16
Cutting Strategy for a Pair of Palettes
Input Palettes
Sorted Palettes
Merged Palettes
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Method #2
SORTING A PAIR OF PALETTES
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 17
Input Palettes
Sorted Palettes
SPS Cutting Strategy
PPS Cutting Strategy
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Method #2
SORTING A PAIR OF PALETTES
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 18
Input Palettes PPS Cutting Strategy SPS Cutting Strategy
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Method #2
SORTING A PAIR OF PALETTES
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 19
Palettes with Different Length More than Two Palettes
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Method #3
PALETTE SIMILARITY MEASUREMENT
How much do they look similar?
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 20
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Method #3
PALETTE SIMILARITY MEASUREMENT
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 21
Previous Similarity Measurements
Element-wise Similarity
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Method #3
PALETTE SIMILARITY MEASUREMENT
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 22
Element-wise Similarity Flow-based Similarity
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Method #3
PALETTE SIMILARITY MEASUREMENT
We proposed a Dynamic Closest Color Warping (DCCW).
− Geometric simularity
− Not element-wise, but flow-based measurement
− Regarding a palette as a sequence, and then measure
the distance between two sequences
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 23
Input Pair Sorted Pair
Geometric Arrangement
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Method #3
PALETTE SIMILARITY MEASUREMENT
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 24
Sort
Input Palettes
Sorted Palettes
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Method #3
PALETTE SIMILARITY MEASUREMENT
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 25
Sorted Palettes
𝐷𝐷𝐶𝐶𝑊 𝐴, 𝐵 =
𝑊 𝐴, 𝐵 + 𝑊(𝐵, 𝐴)
𝐴 + 𝐵
𝑊 𝐴, 𝐵 =
𝑎∈𝐴
min
𝑏𝑖∈𝐵
𝑑(𝑎, 𝑏𝑖𝑏𝑖+1)
A
B
𝒂𝟏 𝒂𝟐 𝒂𝟑 𝒂𝟒 𝒂𝟓 𝒂𝟔 𝒂𝟕 𝒂𝟖 𝒂𝟗 𝒂𝟏𝟎
𝒃𝟏 𝒃𝟐 𝒃𝟑 𝒃𝟒 𝒃𝟓 𝒃𝟔 𝒃𝟕 𝒃𝟖 𝒃𝟗 𝒃𝟏𝟎
𝑑(𝑎, 𝑏𝑖𝑏𝑖+1)
𝒃𝒊+𝟏
𝒃𝒊
𝒂
The shortest Euclidean distance between
CIELAB color 𝑎 and line segment bi𝑏𝑖+1.
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
EVALUATION
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 26
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
EVALUATION OVERVIEW
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 27
Single Palette Sorting
SPS
Palette Pair Sorting
PPS
Similarity Measurement
DCCW
 How naturally the colors are aligned
 How well the colors are aligned between palettes
 How naturally the colors are aligned in each palette
 How effectively DCCW rates similar palettes as similar
 How effectively DCCW rates dissimilar palettes as dissimilar
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Evaluation #1
SINGLE PALETTE SORTING
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 28
Single Palette Sorting
SPS
 How naturally the colors are aligned
If there is a correctly ordered palette,
we can evaluate how well SPS aligned a shuffled palette comparing to a ground-truth palette.
Ground Truth
(Correctly Ordered)
Sorted Palette
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 29
Farnsworth-Munsell 100 Hue Test
• It is designed to test for color blindness, with 85 different hues.
• It tests the ability to distinguish and arrange minute hue differences under constant value
and chroma.
Dataset #1: Farnsworth-Munsell 100 Hue Test Palette Dataset (FM100P)
Evaluation #1
SINGLE PALETTE SORTING
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Evaluation #1
SINGLE PALETTE SORTING
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 30
Farnsworth-Munsell 100 Hue Test
Dataset #1: Farnsworth-Munsell 100 Hue Test Palette Dataset (FM100P)
Correctly Sorted Palette
(Ground Truth)
We generated several FM100P types with
palette lengths 10, 15, 20, 25, 30, 35, and 40,
each comprising 100 palettes.
Maximum Distance: 5
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 31
📐 How can we measure the accuracy of the sorting result?
Ground
Truth
(GT)
Target
Success = The order of the GT and target palette is identical.
Success Rate = (# of successes) / (# of tests)
1. Success Rate
The longest subsequence length for the sorted elements, where
the subsequences are not necessarily contiguous in the sequence
3. Length of the Longest Increasing Subsequence (LLIS)
Measuring the difference between two sequences.
2. Levenshtein Distance
To measure differences
between two palettes
even if the sorted order
is not perfectly correct.
Evaluation #1
SINGLE PALETTE SORTING: ACCURACY
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Evaluation #1
SINGLE PALETTE SORTING: RESULT
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 32
CIELAB HSV GA FIA SA ACO-50 ACO-10 ACO-2 LKH (Ours)
SR 0.001 0.829 0.201 0.011 0.020 0.733 0.631 0.540 0.789
LD 0.884 0.014 0.760 0.849 0.874 0.562 0.562 0.574 0.563
LLIS 0.488 0.993 0.693 0.623 0.463 0.967 0.954 0.941 0.966
ET 0.001 0.001 0.201 0.129 0.294 4.889 1.035 0.552 0.169
Since FM100P was originally sorted by
hue value, it is expected that HSV
lexicographical sorting performs best
Time Complexity of ACO: 𝑂(𝐼n2 𝑚)
Time Complexity of LKH: 𝑂(𝐼𝑛2.2
)
I: # of iterations, m: the number of ants
SR: Success Rate
LD: Levenshetin Distance
ET: Elapsed Time
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
EVALUATION OVERVIEW
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 33
Single Palette Sorting
SPS
Palette Pair Sorting
PPS
Similarity Measurement
DCCW
 How naturally the colors are aligned
 How well the colors are aligned between palettes
 How naturally the colors are aligned in each palette
 How effectively DCCW rates similar palettes as similar
 How effectively DCCW rates dissimilar palettes as dissimilar
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Evaluation #2
PALETTE PAIR SORTING: DATASET
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 34
Palette Pair Sorting
PPS
 How well the colors are aligned between palettes
 How naturally the colors are aligned in each palette
FM100P for SPS Evaluation
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Evaluation #2
PALETTE PAIR SORTING: DATASET
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 35
hue
tone
Kobayashi Hue & Tone 130 System
• It is designed for color semantics based on the
Munsell color system and ISCC-NBS color
naming system.
Neutral
Dataset #2: Kobayashi Hue & Tone Palette Dataset (KHTP)
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Evaluation #2
PALETTE PAIR SORTING: DATASET
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 36
hue
tone
To avoid bipartite matching, we varied row-grouped
palettes by (1) mixing, (2) jittering, and (3) interpolation.
Original Palette Sampling
Dataset #2: Kobayashi Hue & Tone Palette Dataset (KHTP)
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Mixing changes each palette color into a different color in
the same column with a maximum adjacent index
difference of 2.
Evaluation #2
PALETTE PAIR SORTING: DATASET
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 37
hue
tone
Mixing Colors
Dataset #2: Kobayashi Hue & Tone Palette Dataset (KHTP)
To avoid bipartite matching, we varied row-grouped
palettes by (1) mixing, (2) jittering, and (3) interpolation.
Mixing
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Evaluation #2
PALETTE PAIR SORTING: DATASET
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 38
𝛼 = 5
𝛼 = 10
𝛼 = 15
Dataset #2: Kobayashi Hue & Tone Palette Dataset (KHTP)
To avoid bipartite matching, we varied row-grouped
palettes by (1) mixing, (2) jittering, and (3) interpolation.
Jittering
Jittering adds an arbitrary offset, Ω(−𝛼, 𝛼), to each color
to provide minute color changes.
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Evaluation #2
PALETTE PAIR SORTING: DATASET
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 39
Dataset #2: Kobayashi Hue & Tone Palette Dataset (KHTP)
To avoid bipartite matching, we varied row-grouped
palettes by (1) mixing, (2) jittering, and (3) interpolation.
Interpolation
Interpolation adds additional colors to
the palette to disrupt bipartite matching.
#Interpolation = 1
#Interpolation = 3
#Interpolation = 5
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Evaluation #2
PALETTE PAIR SORTING: ACCURACY
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 40
📐 How can we measure whether a pair of palettes are correctly sorted?
Ground Truth Pair
Sorted Pair
A1 A2 A3 A4 A5 A6 A7 A8 A9 A10 A11
B1 B2 B3 B4 B5 B6 B7 B8 B9 B10 B11
A3 A7 A6 A9 A2 A5 A8 A10 A4 A11 A1
B9 B3 B5 B6 B11 B1 B7 B4 B8 B2 B10
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Evaluation #2
PALETTE PAIR SORTING: ACCURACY
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 41
 (Concurrency) How well the colors are aligned between palettes
LLCS indicates the length of the longest subsequence, which
is common to two sequences, and the subsequence is not
necessarily consecutive.
2. Length of the longest common subsequence (LLCS)
Measuring the difference between two sequences.
1. Levenshtein Distance
Sorted
Pair
A3 A7 A6 A9 A2 A5 A8 A10 A4 A11 A1
B9 B3 B5 B6 B11 B1 B7 B4 B8 B2 B10
📐 How can we measure whether a pair of palettes are correctly sorted?
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Evaluation #2
PALETTE PAIR SORTING: ACCURACY
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 42
 (Naturalness) How naturally the colors were aligned in each palette
Sorted A3 A7 A6 A9 A2 A5 A8 A10 A4 A11 A1
B9 B3 B5 B6 B11 B1 B7 B4 B8 B2 B10
Same metric with SPS
4. LLIS of Each Palette
Measuring the difference between sorted and ground truth
palette
3. Levenshtein Distance of Each Palette
Ground Truth
A1 A2 A3 A4 A5 A6 A7 A8 A9 A10 A11
B1 B2 B3 B4 B5 B6 B7 B8 B9 B10 B11
Sorted
Ground Truth
📐 How can we measure whether a pair of palettes are correctly sorted?
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Evaluation #2
PALETTE PAIR SORTING: RESULT
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 43
Ours BPS I-BPS
Naturalness LD 0.631 0.783 0.651
LLIS 0.486 0.455 0.475
Concurrency LD 0.375 0.603 0.430
LLCS 0.769 0.560 0.738
Accuracy of the Palette Pair Sorting
LD: Levenshetin Distance
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Evaluation #2
PALETTE PAIR SORTING: RESULT
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 44
w/o cutting strategy w/ cutting strategy
Δ𝑚𝑎𝑥 Δ𝑎𝑣𝑔 Δ𝑚𝑎𝑥 Δ𝑎𝑣𝑔
w/o jittering and interpolation 27.947 12.767 26.902 12.743
w/ jittering only 29.285 16.529 28.463 16.465
w/ interpolation only 28.438 13.076 27.561 13.071
w/ jittering and interpolation 29.015 16.346 27.901 16.257
𝛥𝑎𝑣𝑔: the average difference of adjacent colors
𝛥𝑚𝑎𝑥: the maximum color difference
Effects of the Cutting Strategy
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
EVALUATION OVERVIEW
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 45
Single Palette Sorting
SPS
Palette Pair Sorting
PPS
Similarity Measurement
DCCW
 How naturally the colors are aligned
 How well the colors are aligned between palettes
 How naturally the colors are aligned in each palette
 How effectively DCCW rates similar palettes as similar
 How effectively DCCW rates dissimilar palettes as dissimilar
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Evaluation #3
DCCW: DATASET
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 46
Similarity Measurement
DCCW
 How effectively DCCW rates similar palettes as similar
 How effectively DCCW rates dissimilar palettes as dissimilar
FM100P for SPS Evaluation
KHTP for PPS Evaluation
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Evaluation #3
DCCW: DATASET
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 47
Lin and Hanrahan’s Palettes Dataset
• [Lin and Hanrahan 2013] generated a color palette dataset to model human’s color palette
recognition from images.
• The dataset comprises unsorted palettes extracted from 40 images.
• by humans including artists and Amazon Mechanical Turk’s participants
• by machines using k-means, c-means clustering, and their extraction model.
Dataset #3: Lin and Hanrahan Similarity Palettes (LHSP)
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Evaluation #3
DCCW: DATASET
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 47
Lin and Hanrahan’s Palettes Dataset
• [Lin and Hanrahan 2013] generated a color palette dataset to model human’s color palette
recognition from images.
• The dataset comprises unsorted palettes extracted from 40 images.
• by humans including artists and Amazon Mechanical Turk’s participants
• by machines using k-means, c-means clustering, and their extraction model.
Dataset #3: Lin and Hanrahan Similarity Palettes (LHSP)
Their
Models Amazon MTurk Artists
Source
Colors
Source Image
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Evaluation #3
DCCW: DATASET
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 47
Lin and Hanrahan’s Palettes Dataset
• [Lin and Hanrahan 2013] generated a color palette dataset to model human’s color palette
recognition from images.
• The dataset comprises unsorted palettes extracted from 40 images.
• by humans including artists and Amazon Mechanical Turk’s participants
• by machines using k-means, c-means clustering, and their extraction model.
Dataset #3: Lin and Hanrahan Similarity Palettes (LHSP)
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Evaluation #3
DCCW: DATASET
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 48
• To make palettes composed of similar but different colors
• Replacement: replaced 10% of the colors of the palette with arbitrary colors.
• Jittering: jittered the colors with jittering 𝛼 ranges = 0, 5, 10, 15
Source Image Example of Generated Palettes
Dataset #3: Lin and Hanrahan Similarity Palettes (LHSP)
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Evaluation #3
DCCW: DATASET
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 49
• We divided the palettes from the same image into disjoint query and retrieval sets.
• We generated several LHSP types with palette lengths 5, 10, 15, 20, 25, and 30.
• LHSP with palette lengths 10~30:
• 50 query palettes (from 10 images)
• 400 retrieval palettes (from 40 images including 10 query images)
• LHSP with palette length 5
• 10 query palettes (from 10 images)
• 1,600 retrieval palettes (from 40 images including 10 query images)
Dataset #3: Lin and Hanrahan Similarity Palettes (LHSP)
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Evaluation #3
DCCW: RESULT
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 50
📐 How well similar palettes from the same image are retrieved among the dataset.
DCCW MPD HD MHD LTS-HD MCD MPHSM CEMD MBME DTW
w/o jittering and
replacement
0.888 0.219 0.571 0.874 0.888 0.880 0.016 0.806 0.813 0.811
w/ jittering only 0.785 0.206 0.524 0.771 0.701 0.763 0.015 0.713 0.713 0.712
w/ replacement only 0.857 0.208 0.206 0.845 0.874 0.857 0.016 0.777 0.788 0.784
w/ jittering and
replacement
0.674 0.189 0.197 0.662 0.631 0.667 0.016 0.630 0.631 0.630
Mean Average Precision (mAP) for LHSP
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Evaluation #3
DCCW: RESULT
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 51
DCCW MPD HD MHD LTS-HD MCD MPHSM CEMD MBME DTW
𝑘 = 5 0.618 0.152 0.345 0.571 0.509 0.588 0.017 0.454 0.457 0.418
𝑘 = 10 0.549 0.176 0.304 0.529 0.484 0.522 0.016 0.429 0.432 0.447
𝑘 = 15 0.683 0.189 0.293 0.678 0.654 0.671 0.015 0.626 0.629 0.632
𝑘 = 20 0.862 0.217 0.371 0.849 0.828 0.848 0.015 0.801 0.804 0.818
𝑘 = 25 0.917 0.237 0.418 0.923 0.904 0.923 0.015 0.924 0.925 0.924
𝑘 = 30 0.963 0.238 0.474 0.963 0.939 0.966 0.015 0.976 0.976 0.976
Mean Average Precision (mAP) for LHSP according to different palette size 𝑘
📐 How well similar palettes from the same image are retrieved among the dataset.
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Evaluation #4
PERCEPTUAL STUDY
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 52
Ours
 Verifying that our method is superior for palette sorting
Purpose
Method  Amazon Mechanical Turk
Questionnaire SPS
Q. Which one do you think is arranged in the most natural order?
HSV-based Sorting
CIELAB-based Sorting
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Evaluation #4
PERCEPTUAL STUDY
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 53
Ours
 Verifying that our method is superior for palette sorting
Purpose
Method  Amazon Mechanical Turk
Questionnaire PPS
Q. Which pair do you think is arranged in the most natural order?
HSV-based Sorting
CIELAB-based Sorting
BPS
I-BPS
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Evaluation #4
PERCEPTUAL STUDY: RESULT
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 54
𝑘 = 10 𝑘 = 15 𝑘 = 20
SPS Ours > HSV-based sorting 79.9% 85.9% 86.7%
Ours > CIELAB-based sorting 69.3% 63.9% 64.4%
PPS Ours > HSV-based sorting 82.9% 88.8% 92.1%
Ours > CIELAB-based sorting 64.1% 60.9% 59.8%
Ours > BPS 86.5% 89.0% 91.4%
Ours > I-BPS 67.4% 64.9% 71.0%
* The effects of palette length and sorting methods were determined using a two-way repeated
measures ANOVA with a significance level of 𝑝 < 0.001 for both SPS and PPS results
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
APPLICATIONS
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 55
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Application #1
PALETTE INTERPOLATION
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 56
HSV-based
CIELAB-based
DCCW-based
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Application #2
PALETTE NAVIGATION
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 57
Kiki's Delivery Service
(1989)
Swallow (2019)
Call Me by Your Name
(2017)
The Place Beyond
the Pines (2012)
Pan's Labyrinth (2006)
Dunkirk (2017)
Query Palette Unsplash
WikiArt
COLOURlovers Commercial Movie
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Django Unchained (2012)
Swallow (2019)
The Master (2012)
The Neon Demon (2016)
Titanic (1997)
World War Z (2013)
Application #2
PALETTE NAVIGATION
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 58
Query Palette Unsplash
WikiArt
COLOURlovers Commercial Movie
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Application #3
PALETTE-BASED RECOLORING
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 59
Recoloring
Convex Hull
Deformation
Original Image
Extracted Palette
Convex Hull
Target Palette Simple Transformation Extending [Tan et al. 2018]
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Application #3
PALETTE-BASED RECOLORING
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 60
Original Image
Target Palette
Improved Transformation
Recoloring Result
Convex Hull
Deformation
Palette Pair Sorting
Source (Extracted)
Target
1
Sorted Source
Sorted target
2
Nearest Matching Using DCCW
Closest Colors
from Target to Source
3
Sorted Target
4
Buffered Palette
Merging
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
Application #3
PALETTE-BASED RECOLORING
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 61
Target Palette
Simple
Transformation
Improved
Transformation
Original Image Target Palette
Simple
Transformation
Improved
Transformation
Original Image
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
CONCLUSION
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 62
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
CONCLUSION
• We proposed a method to sort palette colors by traversing color graphs.
• We proposed a dynamic closest color warping (DCCW) similarity measurement for palette comparison.
• We evaluated our palette sorting and DCCW on various datasets using various criteria confirming ours
outperformed previous methods in terms of accuracy and computation time.
• We also evaluated our methods using a perceptual study, and the result clearly indicates that ours generate
perceptually natural sorting results, outperforming previous techniques.
• We demonstrate many compelling applications enabled by DCCW, including palette interpolation, palette
navigation, and image recoloring.
• Source code and dataset are available at:
• https://github.com/SuziKim/DCCW / https://github.com/SuziKim/DCCW-dataset
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 63
THE PREMIER CONFERENCE & EXHIBITION IN
COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
© 2021 SIGGRAPH. ALL RIGHTS RESERVED. 64
THANK YOU!
SUZI KIM AND SUNGHEE CHOI
GEOMETRIC COMPUTING LAB.
SCHOOL OF COMPUTING, KAIST
DYNAMIC CLOSEST COLOR WARPING
TO SORT AND COMPARE PALETTES

More Related Content

What's hot

汎用なNeural Network Potential「Matlantis」を使った新素材探索_2022応用物理学会_2022/3/22
汎用なNeural Network Potential「Matlantis」を使った新素材探索_2022応用物理学会_2022/3/22汎用なNeural Network Potential「Matlantis」を使った新素材探索_2022応用物理学会_2022/3/22
汎用なNeural Network Potential「Matlantis」を使った新素材探索_2022応用物理学会_2022/3/22Matlantis
 
20160417dlibによる顔器官検出
20160417dlibによる顔器官検出20160417dlibによる顔器官検出
20160417dlibによる顔器官検出Takuya Minagawa
 
久しぶりにMicrosoft Meshを使ってみた - 色々変わってたよ編 -
久しぶりにMicrosoft Meshを使ってみた - 色々変わってたよ編 -久しぶりにMicrosoft Meshを使ってみた - 色々変わってたよ編 -
久しぶりにMicrosoft Meshを使ってみた - 色々変わってたよ編 -Takahiro Miyaura
 
統計的音声合成変換と近年の発展
統計的音声合成変換と近年の発展統計的音声合成変換と近年の発展
統計的音声合成変換と近年の発展Shinnosuke Takamichi
 
Deep-Learning-Based Environmental Sound Segmentation - Integration of Sound ...
Deep-Learning-Based  Environmental Sound Segmentation - Integration of Sound ...Deep-Learning-Based  Environmental Sound Segmentation - Integration of Sound ...
Deep-Learning-Based Environmental Sound Segmentation - Integration of Sound ...Yui Sudo
 
Surveyから始まる研究者への道 - Stand on the shoulders of giants -
Surveyから始まる研究者への道 - Stand on the shoulders of giants -Surveyから始まる研究者への道 - Stand on the shoulders of giants -
Surveyから始まる研究者への道 - Stand on the shoulders of giants -諒介 荒木
 
30th コンピュータビジョン勉強会@関東 DynamicFusion
30th コンピュータビジョン勉強会@関東 DynamicFusion30th コンピュータビジョン勉強会@関東 DynamicFusion
30th コンピュータビジョン勉強会@関東 DynamicFusionHiroki Mizuno
 
信号の独立性に基づく多チャンネル音源分離
信号の独立性に基づく多チャンネル音源分離信号の独立性に基づく多チャンネル音源分離
信号の独立性に基づく多チャンネル音源分離NU_I_TODALAB
 
[DL輪読会]Wav2CLIP: Learning Robust Audio Representations From CLIP
[DL輪読会]Wav2CLIP: Learning Robust Audio Representations From CLIP[DL輪読会]Wav2CLIP: Learning Robust Audio Representations From CLIP
[DL輪読会]Wav2CLIP: Learning Robust Audio Representations From CLIPDeep Learning JP
 
九州大学教育ビッグデータプロジェクト ~ラーニングアナリティクス(LA)の活用~
九州大学教育ビッグデータプロジェクト ~ラーニングアナリティクス(LA)の活用~九州大学教育ビッグデータプロジェクト ~ラーニングアナリティクス(LA)の活用~
九州大学教育ビッグデータプロジェクト ~ラーニングアナリティクス(LA)の活用~Japan Electronic Publishing Association
 
remote Docker over SSHが熱い
remote Docker over SSHが熱いremote Docker over SSHが熱い
remote Docker over SSHが熱いHiroyuki Ohnaka
 
CNNの構造最適化手法(第3回3D勉強会)
CNNの構造最適化手法(第3回3D勉強会)CNNの構造最適化手法(第3回3D勉強会)
CNNの構造最適化手法(第3回3D勉強会)MasanoriSuganuma
 
文献紹介:Simpler Is Better: Few-Shot Semantic Segmentation With Classifier Weight...
文献紹介:Simpler Is Better: Few-Shot Semantic Segmentation With Classifier Weight...文献紹介:Simpler Is Better: Few-Shot Semantic Segmentation With Classifier Weight...
文献紹介:Simpler Is Better: Few-Shot Semantic Segmentation With Classifier Weight...Toru Tamaki
 
超簡単でハイセンスな表紙スライドの作り方
超簡単でハイセンスな表紙スライドの作り方超簡単でハイセンスな表紙スライドの作り方
超簡単でハイセンスな表紙スライドの作り方MOCKS | Yuta Morishige
 
ニューラルネットと深層学習の歴史
ニューラルネットと深層学習の歴史ニューラルネットと深層学習の歴史
ニューラルネットと深層学習の歴史Akinori Abe
 
自然言語処理を 役立てるのはなぜ難しいのか(2022/10/25東大大学院「自然言語処理応用」)
自然言語処理を 役立てるのはなぜ難しいのか(2022/10/25東大大学院「自然言語処理応用」)自然言語処理を 役立てるのはなぜ難しいのか(2022/10/25東大大学院「自然言語処理応用」)
自然言語処理を 役立てるのはなぜ難しいのか(2022/10/25東大大学院「自然言語処理応用」)Preferred Networks
 
HigherHRNet: Scale-Aware Representation Learning for Bottom-Up Human Pose Est...
HigherHRNet: Scale-Aware Representation Learning for Bottom-Up Human Pose Est...HigherHRNet: Scale-Aware Representation Learning for Bottom-Up Human Pose Est...
HigherHRNet: Scale-Aware Representation Learning for Bottom-Up Human Pose Est...harmonylab
 
NeurIPS2021論文読み会 (parameter prediction for unseen deep architecture)
NeurIPS2021論文読み会 (parameter prediction for unseen deep architecture)NeurIPS2021論文読み会 (parameter prediction for unseen deep architecture)
NeurIPS2021論文読み会 (parameter prediction for unseen deep architecture)Kshinmei
 
畳み込みニューラルネットワークの研究動向
畳み込みニューラルネットワークの研究動向畳み込みニューラルネットワークの研究動向
畳み込みニューラルネットワークの研究動向Yusuke Uchida
 
深層学習による製造業のスマート化と産業応用の将来展望(クオリティフォーラム2020講演資料)
深層学習による製造業のスマート化と産業応用の将来展望(クオリティフォーラム2020講演資料)深層学習による製造業のスマート化と産業応用の将来展望(クオリティフォーラム2020講演資料)
深層学習による製造業のスマート化と産業応用の将来展望(クオリティフォーラム2020講演資料)Preferred Networks
 

What's hot (20)

汎用なNeural Network Potential「Matlantis」を使った新素材探索_2022応用物理学会_2022/3/22
汎用なNeural Network Potential「Matlantis」を使った新素材探索_2022応用物理学会_2022/3/22汎用なNeural Network Potential「Matlantis」を使った新素材探索_2022応用物理学会_2022/3/22
汎用なNeural Network Potential「Matlantis」を使った新素材探索_2022応用物理学会_2022/3/22
 
20160417dlibによる顔器官検出
20160417dlibによる顔器官検出20160417dlibによる顔器官検出
20160417dlibによる顔器官検出
 
久しぶりにMicrosoft Meshを使ってみた - 色々変わってたよ編 -
久しぶりにMicrosoft Meshを使ってみた - 色々変わってたよ編 -久しぶりにMicrosoft Meshを使ってみた - 色々変わってたよ編 -
久しぶりにMicrosoft Meshを使ってみた - 色々変わってたよ編 -
 
統計的音声合成変換と近年の発展
統計的音声合成変換と近年の発展統計的音声合成変換と近年の発展
統計的音声合成変換と近年の発展
 
Deep-Learning-Based Environmental Sound Segmentation - Integration of Sound ...
Deep-Learning-Based  Environmental Sound Segmentation - Integration of Sound ...Deep-Learning-Based  Environmental Sound Segmentation - Integration of Sound ...
Deep-Learning-Based Environmental Sound Segmentation - Integration of Sound ...
 
Surveyから始まる研究者への道 - Stand on the shoulders of giants -
Surveyから始まる研究者への道 - Stand on the shoulders of giants -Surveyから始まる研究者への道 - Stand on the shoulders of giants -
Surveyから始まる研究者への道 - Stand on the shoulders of giants -
 
30th コンピュータビジョン勉強会@関東 DynamicFusion
30th コンピュータビジョン勉強会@関東 DynamicFusion30th コンピュータビジョン勉強会@関東 DynamicFusion
30th コンピュータビジョン勉強会@関東 DynamicFusion
 
信号の独立性に基づく多チャンネル音源分離
信号の独立性に基づく多チャンネル音源分離信号の独立性に基づく多チャンネル音源分離
信号の独立性に基づく多チャンネル音源分離
 
[DL輪読会]Wav2CLIP: Learning Robust Audio Representations From CLIP
[DL輪読会]Wav2CLIP: Learning Robust Audio Representations From CLIP[DL輪読会]Wav2CLIP: Learning Robust Audio Representations From CLIP
[DL輪読会]Wav2CLIP: Learning Robust Audio Representations From CLIP
 
九州大学教育ビッグデータプロジェクト ~ラーニングアナリティクス(LA)の活用~
九州大学教育ビッグデータプロジェクト ~ラーニングアナリティクス(LA)の活用~九州大学教育ビッグデータプロジェクト ~ラーニングアナリティクス(LA)の活用~
九州大学教育ビッグデータプロジェクト ~ラーニングアナリティクス(LA)の活用~
 
remote Docker over SSHが熱い
remote Docker over SSHが熱いremote Docker over SSHが熱い
remote Docker over SSHが熱い
 
CNNの構造最適化手法(第3回3D勉強会)
CNNの構造最適化手法(第3回3D勉強会)CNNの構造最適化手法(第3回3D勉強会)
CNNの構造最適化手法(第3回3D勉強会)
 
文献紹介:Simpler Is Better: Few-Shot Semantic Segmentation With Classifier Weight...
文献紹介:Simpler Is Better: Few-Shot Semantic Segmentation With Classifier Weight...文献紹介:Simpler Is Better: Few-Shot Semantic Segmentation With Classifier Weight...
文献紹介:Simpler Is Better: Few-Shot Semantic Segmentation With Classifier Weight...
 
超簡単でハイセンスな表紙スライドの作り方
超簡単でハイセンスな表紙スライドの作り方超簡単でハイセンスな表紙スライドの作り方
超簡単でハイセンスな表紙スライドの作り方
 
ニューラルネットと深層学習の歴史
ニューラルネットと深層学習の歴史ニューラルネットと深層学習の歴史
ニューラルネットと深層学習の歴史
 
自然言語処理を 役立てるのはなぜ難しいのか(2022/10/25東大大学院「自然言語処理応用」)
自然言語処理を 役立てるのはなぜ難しいのか(2022/10/25東大大学院「自然言語処理応用」)自然言語処理を 役立てるのはなぜ難しいのか(2022/10/25東大大学院「自然言語処理応用」)
自然言語処理を 役立てるのはなぜ難しいのか(2022/10/25東大大学院「自然言語処理応用」)
 
HigherHRNet: Scale-Aware Representation Learning for Bottom-Up Human Pose Est...
HigherHRNet: Scale-Aware Representation Learning for Bottom-Up Human Pose Est...HigherHRNet: Scale-Aware Representation Learning for Bottom-Up Human Pose Est...
HigherHRNet: Scale-Aware Representation Learning for Bottom-Up Human Pose Est...
 
NeurIPS2021論文読み会 (parameter prediction for unseen deep architecture)
NeurIPS2021論文読み会 (parameter prediction for unseen deep architecture)NeurIPS2021論文読み会 (parameter prediction for unseen deep architecture)
NeurIPS2021論文読み会 (parameter prediction for unseen deep architecture)
 
畳み込みニューラルネットワークの研究動向
畳み込みニューラルネットワークの研究動向畳み込みニューラルネットワークの研究動向
畳み込みニューラルネットワークの研究動向
 
深層学習による製造業のスマート化と産業応用の将来展望(クオリティフォーラム2020講演資料)
深層学習による製造業のスマート化と産業応用の将来展望(クオリティフォーラム2020講演資料)深層学習による製造業のスマート化と産業応用の将来展望(クオリティフォーラム2020講演資料)
深層学習による製造業のスマート化と産業応用の将来展望(クオリティフォーラム2020講演資料)
 

Similar to [SIGGRAPH 2021] Dynamic Closest Color Warping to Sort and Compare Palettes

Half-Life_Alyx_Locomotion_Slides.pdf
Half-Life_Alyx_Locomotion_Slides.pdfHalf-Life_Alyx_Locomotion_Slides.pdf
Half-Life_Alyx_Locomotion_Slides.pdfLiJimmy1
 
Human Perception and Colour Theory - Lecture 2 - Information Visualisation (4...
Human Perception and Colour Theory - Lecture 2 - Information Visualisation (4...Human Perception and Colour Theory - Lecture 2 - Information Visualisation (4...
Human Perception and Colour Theory - Lecture 2 - Information Visualisation (4...Beat Signer
 
A N A LTERNATIVE G REEN S CREEN K EYING M ETHOD F OR F ILM V ISUAL E ...
A N  A LTERNATIVE  G REEN  S CREEN  K EYING M ETHOD  F OR  F ILM  V ISUAL  E ...A N  A LTERNATIVE  G REEN  S CREEN  K EYING M ETHOD  F OR  F ILM  V ISUAL  E ...
A N A LTERNATIVE G REEN S CREEN K EYING M ETHOD F OR F ILM V ISUAL E ...ijma
 
A Comparative Study of Different Models on Image Colorization using Deep Lear...
A Comparative Study of Different Models on Image Colorization using Deep Lear...A Comparative Study of Different Models on Image Colorization using Deep Lear...
A Comparative Study of Different Models on Image Colorization using Deep Lear...IRJET Journal
 
Graph Coloring and Its Implementation
Graph Coloring and Its ImplementationGraph Coloring and Its Implementation
Graph Coloring and Its ImplementationIJARIIT
 
Pbj Painting | Painting a house
Pbj Painting | Painting a house Pbj Painting | Painting a house
Pbj Painting | Painting a house pbjpainting
 
PLAN601E SESSION 2 LESSON
PLAN601E SESSION 2 LESSONPLAN601E SESSION 2 LESSON
PLAN601E SESSION 2 LESSONrkottam
 
Color sorting machine using color light to frequency converter
Color sorting machine using color light to frequency converterColor sorting machine using color light to frequency converter
Color sorting machine using color light to frequency converterAssiaHAMZA
 
Color image steganography in YCbCr space
Color image steganography in YCbCr spaceColor image steganography in YCbCr space
Color image steganography in YCbCr spaceIJECEIAES
 
Advance DLP for Superior 3D Printing
Advance DLP for Superior 3D Printing Advance DLP for Superior 3D Printing
Advance DLP for Superior 3D Printing Engineering Technique
 
LCD Characterization Report
LCD Characterization ReportLCD Characterization Report
LCD Characterization ReportTanmay Mondal
 
Colorspace: Useful For More Than Just Color? - SF Video Tech Meetup - 27 May ...
Colorspace: Useful For More Than Just Color? - SF Video Tech Meetup - 27 May ...Colorspace: Useful For More Than Just Color? - SF Video Tech Meetup - 27 May ...
Colorspace: Useful For More Than Just Color? - SF Video Tech Meetup - 27 May ...Derek Buitenhuis
 
CS 354 Understanding Color
CS 354 Understanding ColorCS 354 Understanding Color
CS 354 Understanding ColorMark Kilgard
 
Image Maximization Using Multi Spectral Image Fusion Technique
Image Maximization Using Multi Spectral Image Fusion TechniqueImage Maximization Using Multi Spectral Image Fusion Technique
Image Maximization Using Multi Spectral Image Fusion Techniquedbpublications
 

Similar to [SIGGRAPH 2021] Dynamic Closest Color Warping to Sort and Compare Palettes (17)

Half-Life_Alyx_Locomotion_Slides.pdf
Half-Life_Alyx_Locomotion_Slides.pdfHalf-Life_Alyx_Locomotion_Slides.pdf
Half-Life_Alyx_Locomotion_Slides.pdf
 
Human Perception and Colour Theory - Lecture 2 - Information Visualisation (4...
Human Perception and Colour Theory - Lecture 2 - Information Visualisation (4...Human Perception and Colour Theory - Lecture 2 - Information Visualisation (4...
Human Perception and Colour Theory - Lecture 2 - Information Visualisation (4...
 
A N A LTERNATIVE G REEN S CREEN K EYING M ETHOD F OR F ILM V ISUAL E ...
A N  A LTERNATIVE  G REEN  S CREEN  K EYING M ETHOD  F OR  F ILM  V ISUAL  E ...A N  A LTERNATIVE  G REEN  S CREEN  K EYING M ETHOD  F OR  F ILM  V ISUAL  E ...
A N A LTERNATIVE G REEN S CREEN K EYING M ETHOD F OR F ILM V ISUAL E ...
 
HDR and WCG Principles-Part 5
HDR and WCG Principles-Part 5HDR and WCG Principles-Part 5
HDR and WCG Principles-Part 5
 
A Comparative Study of Different Models on Image Colorization using Deep Lear...
A Comparative Study of Different Models on Image Colorization using Deep Lear...A Comparative Study of Different Models on Image Colorization using Deep Lear...
A Comparative Study of Different Models on Image Colorization using Deep Lear...
 
Graph Coloring and Its Implementation
Graph Coloring and Its ImplementationGraph Coloring and Its Implementation
Graph Coloring and Its Implementation
 
Pbj Painting | Painting a house
Pbj Painting | Painting a house Pbj Painting | Painting a house
Pbj Painting | Painting a house
 
PLAN601E SESSION 2 LESSON
PLAN601E SESSION 2 LESSONPLAN601E SESSION 2 LESSON
PLAN601E SESSION 2 LESSON
 
Color sorting machine using color light to frequency converter
Color sorting machine using color light to frequency converterColor sorting machine using color light to frequency converter
Color sorting machine using color light to frequency converter
 
Color image steganography in YCbCr space
Color image steganography in YCbCr spaceColor image steganography in YCbCr space
Color image steganography in YCbCr space
 
Advance DLP for Superior 3D Printing
Advance DLP for Superior 3D Printing Advance DLP for Superior 3D Printing
Advance DLP for Superior 3D Printing
 
LCD Characterization Report
LCD Characterization ReportLCD Characterization Report
LCD Characterization Report
 
Colorspace: Useful For More Than Just Color? - SF Video Tech Meetup - 27 May ...
Colorspace: Useful For More Than Just Color? - SF Video Tech Meetup - 27 May ...Colorspace: Useful For More Than Just Color? - SF Video Tech Meetup - 27 May ...
Colorspace: Useful For More Than Just Color? - SF Video Tech Meetup - 27 May ...
 
G04654247
G04654247G04654247
G04654247
 
CS 354 Understanding Color
CS 354 Understanding ColorCS 354 Understanding Color
CS 354 Understanding Color
 
Image Maximization Using Multi Spectral Image Fusion Technique
Image Maximization Using Multi Spectral Image Fusion TechniqueImage Maximization Using Multi Spectral Image Fusion Technique
Image Maximization Using Multi Spectral Image Fusion Technique
 
02 Mm1 B03
02 Mm1 B0302 Mm1 B03
02 Mm1 B03
 

More from Suzi Kim

[KCC 2020] 군집 기반 색상 팔레트 비교
[KCC 2020] 군집 기반 색상 팔레트 비교[KCC 2020] 군집 기반 색상 팔레트 비교
[KCC 2020] 군집 기반 색상 팔레트 비교Suzi Kim
 
[Eyewear 2019] The Hitchhiker’s Guide to the Eyewear Applications
[Eyewear 2019] The Hitchhiker’s Guide to the Eyewear Applications[Eyewear 2019] The Hitchhiker’s Guide to the Eyewear Applications
[Eyewear 2019] The Hitchhiker’s Guide to the Eyewear ApplicationsSuzi Kim
 
[ICMR 2020] Automatic Color Scheme Extraction from Movies
[ICMR 2020] Automatic Color Scheme Extraction from Movies[ICMR 2020] Automatic Color Scheme Extraction from Movies
[ICMR 2020] Automatic Color Scheme Extraction from MoviesSuzi Kim
 
[KSC 2019] 이미지 분할 및 관심 영역 기반 색상 스키마 추출 (Color Scheme Extraction Using Segment...
[KSC 2019] 이미지 분할 및 관심 영역 기반 색상 스키마 추출 (Color Scheme Extraction Using Segment...[KSC 2019] 이미지 분할 및 관심 영역 기반 색상 스키마 추출 (Color Scheme Extraction Using Segment...
[KSC 2019] 이미지 분할 및 관심 영역 기반 색상 스키마 추출 (Color Scheme Extraction Using Segment...Suzi Kim
 
[EMBC 2018] Automatic tooth segmentation of dental mesh using a transverse plane
[EMBC 2018] Automatic tooth segmentation of dental mesh using a transverse plane[EMBC 2018] Automatic tooth segmentation of dental mesh using a transverse plane
[EMBC 2018] Automatic tooth segmentation of dental mesh using a transverse planeSuzi Kim
 
[KCC 2019] CNN 기반 물체 파지를 위한 위치 탐색 (CNN-based Grasping Box Detection)
[KCC 2019] CNN 기반 물체 파지를 위한 위치 탐색 (CNN-based Grasping Box Detection)[KCC 2019] CNN 기반 물체 파지를 위한 위치 탐색 (CNN-based Grasping Box Detection)
[KCC 2019] CNN 기반 물체 파지를 위한 위치 탐색 (CNN-based Grasping Box Detection)Suzi Kim
 
[KSC 2018] 관심 영역 기반 픽셀 아트 이미지 생성 (Saliency-based Pixel Art Image Generation)
[KSC 2018] 관심 영역 기반 픽셀 아트 이미지 생성 (Saliency-based Pixel Art Image Generation)[KSC 2018] 관심 영역 기반 픽셀 아트 이미지 생성 (Saliency-based Pixel Art Image Generation)
[KSC 2018] 관심 영역 기반 픽셀 아트 이미지 생성 (Saliency-based Pixel Art Image Generation)Suzi Kim
 
[KCC 2018] 관심 영역 기반 이미지 자리 표시자 생성 (Saliency-based SVG Image Placeholder Gener...
[KCC 2018] 관심 영역 기반 이미지 자리 표시자 생성 (Saliency-based SVG Image Placeholder Gener...[KCC 2018] 관심 영역 기반 이미지 자리 표시자 생성 (Saliency-based SVG Image Placeholder Gener...
[KCC 2018] 관심 영역 기반 이미지 자리 표시자 생성 (Saliency-based SVG Image Placeholder Gener...Suzi Kim
 
[ICWE 2019] DotCHA: A 3D Text-Based Scatter-Type CAPTCHA (20190612)
[ICWE 2019] DotCHA: A 3D Text-Based Scatter-Type CAPTCHA (20190612)[ICWE 2019] DotCHA: A 3D Text-Based Scatter-Type CAPTCHA (20190612)
[ICWE 2019] DotCHA: A 3D Text-Based Scatter-Type CAPTCHA (20190612)Suzi Kim
 

More from Suzi Kim (9)

[KCC 2020] 군집 기반 색상 팔레트 비교
[KCC 2020] 군집 기반 색상 팔레트 비교[KCC 2020] 군집 기반 색상 팔레트 비교
[KCC 2020] 군집 기반 색상 팔레트 비교
 
[Eyewear 2019] The Hitchhiker’s Guide to the Eyewear Applications
[Eyewear 2019] The Hitchhiker’s Guide to the Eyewear Applications[Eyewear 2019] The Hitchhiker’s Guide to the Eyewear Applications
[Eyewear 2019] The Hitchhiker’s Guide to the Eyewear Applications
 
[ICMR 2020] Automatic Color Scheme Extraction from Movies
[ICMR 2020] Automatic Color Scheme Extraction from Movies[ICMR 2020] Automatic Color Scheme Extraction from Movies
[ICMR 2020] Automatic Color Scheme Extraction from Movies
 
[KSC 2019] 이미지 분할 및 관심 영역 기반 색상 스키마 추출 (Color Scheme Extraction Using Segment...
[KSC 2019] 이미지 분할 및 관심 영역 기반 색상 스키마 추출 (Color Scheme Extraction Using Segment...[KSC 2019] 이미지 분할 및 관심 영역 기반 색상 스키마 추출 (Color Scheme Extraction Using Segment...
[KSC 2019] 이미지 분할 및 관심 영역 기반 색상 스키마 추출 (Color Scheme Extraction Using Segment...
 
[EMBC 2018] Automatic tooth segmentation of dental mesh using a transverse plane
[EMBC 2018] Automatic tooth segmentation of dental mesh using a transverse plane[EMBC 2018] Automatic tooth segmentation of dental mesh using a transverse plane
[EMBC 2018] Automatic tooth segmentation of dental mesh using a transverse plane
 
[KCC 2019] CNN 기반 물체 파지를 위한 위치 탐색 (CNN-based Grasping Box Detection)
[KCC 2019] CNN 기반 물체 파지를 위한 위치 탐색 (CNN-based Grasping Box Detection)[KCC 2019] CNN 기반 물체 파지를 위한 위치 탐색 (CNN-based Grasping Box Detection)
[KCC 2019] CNN 기반 물체 파지를 위한 위치 탐색 (CNN-based Grasping Box Detection)
 
[KSC 2018] 관심 영역 기반 픽셀 아트 이미지 생성 (Saliency-based Pixel Art Image Generation)
[KSC 2018] 관심 영역 기반 픽셀 아트 이미지 생성 (Saliency-based Pixel Art Image Generation)[KSC 2018] 관심 영역 기반 픽셀 아트 이미지 생성 (Saliency-based Pixel Art Image Generation)
[KSC 2018] 관심 영역 기반 픽셀 아트 이미지 생성 (Saliency-based Pixel Art Image Generation)
 
[KCC 2018] 관심 영역 기반 이미지 자리 표시자 생성 (Saliency-based SVG Image Placeholder Gener...
[KCC 2018] 관심 영역 기반 이미지 자리 표시자 생성 (Saliency-based SVG Image Placeholder Gener...[KCC 2018] 관심 영역 기반 이미지 자리 표시자 생성 (Saliency-based SVG Image Placeholder Gener...
[KCC 2018] 관심 영역 기반 이미지 자리 표시자 생성 (Saliency-based SVG Image Placeholder Gener...
 
[ICWE 2019] DotCHA: A 3D Text-Based Scatter-Type CAPTCHA (20190612)
[ICWE 2019] DotCHA: A 3D Text-Based Scatter-Type CAPTCHA (20190612)[ICWE 2019] DotCHA: A 3D Text-Based Scatter-Type CAPTCHA (20190612)
[ICWE 2019] DotCHA: A 3D Text-Based Scatter-Type CAPTCHA (20190612)
 

Recently uploaded

KIT-601 Lecture Notes-UNIT-5.pdf Frame Works and Visualization
KIT-601 Lecture Notes-UNIT-5.pdf Frame Works and VisualizationKIT-601 Lecture Notes-UNIT-5.pdf Frame Works and Visualization
KIT-601 Lecture Notes-UNIT-5.pdf Frame Works and VisualizationDr. Radhey Shyam
 
ASME IX(9) 2007 Full Version .pdf
ASME IX(9)  2007 Full Version       .pdfASME IX(9)  2007 Full Version       .pdf
ASME IX(9) 2007 Full Version .pdfAhmedHussein950959
 
Quality defects in TMT Bars, Possible causes and Potential Solutions.
Quality defects in TMT Bars, Possible causes and Potential Solutions.Quality defects in TMT Bars, Possible causes and Potential Solutions.
Quality defects in TMT Bars, Possible causes and Potential Solutions.PrashantGoswami42
 
Furniture showroom management system project.pdf
Furniture showroom management system project.pdfFurniture showroom management system project.pdf
Furniture showroom management system project.pdfKamal Acharya
 
RS Khurmi Machine Design Clutch and Brake Exercise Numerical Solutions
RS Khurmi Machine Design Clutch and Brake Exercise Numerical SolutionsRS Khurmi Machine Design Clutch and Brake Exercise Numerical Solutions
RS Khurmi Machine Design Clutch and Brake Exercise Numerical SolutionsAtif Razi
 
Immunizing Image Classifiers Against Localized Adversary Attacks
Immunizing Image Classifiers Against Localized Adversary AttacksImmunizing Image Classifiers Against Localized Adversary Attacks
Immunizing Image Classifiers Against Localized Adversary Attacksgerogepatton
 
fundamentals of drawing and isometric and orthographic projection
fundamentals of drawing and isometric and orthographic projectionfundamentals of drawing and isometric and orthographic projection
fundamentals of drawing and isometric and orthographic projectionjeevanprasad8
 
Automobile Management System Project Report.pdf
Automobile Management System Project Report.pdfAutomobile Management System Project Report.pdf
Automobile Management System Project Report.pdfKamal Acharya
 
NO1 Pandit Amil Baba In Bahawalpur, Sargodha, Sialkot, Sheikhupura, Rahim Yar...
NO1 Pandit Amil Baba In Bahawalpur, Sargodha, Sialkot, Sheikhupura, Rahim Yar...NO1 Pandit Amil Baba In Bahawalpur, Sargodha, Sialkot, Sheikhupura, Rahim Yar...
NO1 Pandit Amil Baba In Bahawalpur, Sargodha, Sialkot, Sheikhupura, Rahim Yar...Amil baba
 
Laundry management system project report.pdf
Laundry management system project report.pdfLaundry management system project report.pdf
Laundry management system project report.pdfKamal Acharya
 
Arduino based vehicle speed tracker project
Arduino based vehicle speed tracker projectArduino based vehicle speed tracker project
Arduino based vehicle speed tracker projectRased Khan
 
Top 13 Famous Civil Engineering Scientist
Top 13 Famous Civil Engineering ScientistTop 13 Famous Civil Engineering Scientist
Top 13 Famous Civil Engineering Scientistgettygaming1
 
Natalia Rutkowska - BIM School Course in Kraków
Natalia Rutkowska - BIM School Course in KrakówNatalia Rutkowska - BIM School Course in Kraków
Natalia Rutkowska - BIM School Course in Krakówbim.edu.pl
 
KIT-601 Lecture Notes-UNIT-3.pdf Mining Data Stream
KIT-601 Lecture Notes-UNIT-3.pdf Mining Data StreamKIT-601 Lecture Notes-UNIT-3.pdf Mining Data Stream
KIT-601 Lecture Notes-UNIT-3.pdf Mining Data StreamDr. Radhey Shyam
 
2024 DevOps Pro Europe - Growing at the edge
2024 DevOps Pro Europe - Growing at the edge2024 DevOps Pro Europe - Growing at the edge
2024 DevOps Pro Europe - Growing at the edgePaco Orozco
 
İTÜ CAD and Reverse Engineering Workshop
İTÜ CAD and Reverse Engineering WorkshopİTÜ CAD and Reverse Engineering Workshop
İTÜ CAD and Reverse Engineering WorkshopEmre Günaydın
 
Event Management System Vb Net Project Report.pdf
Event Management System Vb Net  Project Report.pdfEvent Management System Vb Net  Project Report.pdf
Event Management System Vb Net Project Report.pdfKamal Acharya
 
HYDROPOWER - Hydroelectric power generation
HYDROPOWER - Hydroelectric power generationHYDROPOWER - Hydroelectric power generation
HYDROPOWER - Hydroelectric power generationRobbie Edward Sayers
 
ONLINE VEHICLE RENTAL SYSTEM PROJECT REPORT.pdf
ONLINE VEHICLE RENTAL SYSTEM PROJECT REPORT.pdfONLINE VEHICLE RENTAL SYSTEM PROJECT REPORT.pdf
ONLINE VEHICLE RENTAL SYSTEM PROJECT REPORT.pdfKamal Acharya
 
The Benefits and Techniques of Trenchless Pipe Repair.pdf
The Benefits and Techniques of Trenchless Pipe Repair.pdfThe Benefits and Techniques of Trenchless Pipe Repair.pdf
The Benefits and Techniques of Trenchless Pipe Repair.pdfPipe Restoration Solutions
 

Recently uploaded (20)

KIT-601 Lecture Notes-UNIT-5.pdf Frame Works and Visualization
KIT-601 Lecture Notes-UNIT-5.pdf Frame Works and VisualizationKIT-601 Lecture Notes-UNIT-5.pdf Frame Works and Visualization
KIT-601 Lecture Notes-UNIT-5.pdf Frame Works and Visualization
 
ASME IX(9) 2007 Full Version .pdf
ASME IX(9)  2007 Full Version       .pdfASME IX(9)  2007 Full Version       .pdf
ASME IX(9) 2007 Full Version .pdf
 
Quality defects in TMT Bars, Possible causes and Potential Solutions.
Quality defects in TMT Bars, Possible causes and Potential Solutions.Quality defects in TMT Bars, Possible causes and Potential Solutions.
Quality defects in TMT Bars, Possible causes and Potential Solutions.
 
Furniture showroom management system project.pdf
Furniture showroom management system project.pdfFurniture showroom management system project.pdf
Furniture showroom management system project.pdf
 
RS Khurmi Machine Design Clutch and Brake Exercise Numerical Solutions
RS Khurmi Machine Design Clutch and Brake Exercise Numerical SolutionsRS Khurmi Machine Design Clutch and Brake Exercise Numerical Solutions
RS Khurmi Machine Design Clutch and Brake Exercise Numerical Solutions
 
Immunizing Image Classifiers Against Localized Adversary Attacks
Immunizing Image Classifiers Against Localized Adversary AttacksImmunizing Image Classifiers Against Localized Adversary Attacks
Immunizing Image Classifiers Against Localized Adversary Attacks
 
fundamentals of drawing and isometric and orthographic projection
fundamentals of drawing and isometric and orthographic projectionfundamentals of drawing and isometric and orthographic projection
fundamentals of drawing and isometric and orthographic projection
 
Automobile Management System Project Report.pdf
Automobile Management System Project Report.pdfAutomobile Management System Project Report.pdf
Automobile Management System Project Report.pdf
 
NO1 Pandit Amil Baba In Bahawalpur, Sargodha, Sialkot, Sheikhupura, Rahim Yar...
NO1 Pandit Amil Baba In Bahawalpur, Sargodha, Sialkot, Sheikhupura, Rahim Yar...NO1 Pandit Amil Baba In Bahawalpur, Sargodha, Sialkot, Sheikhupura, Rahim Yar...
NO1 Pandit Amil Baba In Bahawalpur, Sargodha, Sialkot, Sheikhupura, Rahim Yar...
 
Laundry management system project report.pdf
Laundry management system project report.pdfLaundry management system project report.pdf
Laundry management system project report.pdf
 
Arduino based vehicle speed tracker project
Arduino based vehicle speed tracker projectArduino based vehicle speed tracker project
Arduino based vehicle speed tracker project
 
Top 13 Famous Civil Engineering Scientist
Top 13 Famous Civil Engineering ScientistTop 13 Famous Civil Engineering Scientist
Top 13 Famous Civil Engineering Scientist
 
Natalia Rutkowska - BIM School Course in Kraków
Natalia Rutkowska - BIM School Course in KrakówNatalia Rutkowska - BIM School Course in Kraków
Natalia Rutkowska - BIM School Course in Kraków
 
KIT-601 Lecture Notes-UNIT-3.pdf Mining Data Stream
KIT-601 Lecture Notes-UNIT-3.pdf Mining Data StreamKIT-601 Lecture Notes-UNIT-3.pdf Mining Data Stream
KIT-601 Lecture Notes-UNIT-3.pdf Mining Data Stream
 
2024 DevOps Pro Europe - Growing at the edge
2024 DevOps Pro Europe - Growing at the edge2024 DevOps Pro Europe - Growing at the edge
2024 DevOps Pro Europe - Growing at the edge
 
İTÜ CAD and Reverse Engineering Workshop
İTÜ CAD and Reverse Engineering WorkshopİTÜ CAD and Reverse Engineering Workshop
İTÜ CAD and Reverse Engineering Workshop
 
Event Management System Vb Net Project Report.pdf
Event Management System Vb Net  Project Report.pdfEvent Management System Vb Net  Project Report.pdf
Event Management System Vb Net Project Report.pdf
 
HYDROPOWER - Hydroelectric power generation
HYDROPOWER - Hydroelectric power generationHYDROPOWER - Hydroelectric power generation
HYDROPOWER - Hydroelectric power generation
 
ONLINE VEHICLE RENTAL SYSTEM PROJECT REPORT.pdf
ONLINE VEHICLE RENTAL SYSTEM PROJECT REPORT.pdfONLINE VEHICLE RENTAL SYSTEM PROJECT REPORT.pdf
ONLINE VEHICLE RENTAL SYSTEM PROJECT REPORT.pdf
 
The Benefits and Techniques of Trenchless Pipe Repair.pdf
The Benefits and Techniques of Trenchless Pipe Repair.pdfThe Benefits and Techniques of Trenchless Pipe Repair.pdf
The Benefits and Techniques of Trenchless Pipe Repair.pdf
 

[SIGGRAPH 2021] Dynamic Closest Color Warping to Sort and Compare Palettes

  • 1. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES DYNAMIC CLOSEST COLOR WARPING TO SORT AND COMPARE PALETTES © 2021 SIGGRAPH. ALL RIGHTS RESERVED. SUZI KIM SUNGHEE CHOI
  • 2. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES WHAT IS A COLOR PALETTE? A limited number of colors expressed in refined shapes © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 2
  • 3. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES WHAT IS A COLOR PALETTE? © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 2 Intuitiveness Generality Easy-computation Simplicity A limited number of colors expressed in refined shapes
  • 4. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES WHAT IS A COLOR PALETTE? © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 2 Intuitiveness Generality Easy-computation Simplicity A limited number of colors expressed in refined shapes
  • 5. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES SORT THESE COLORS! © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 3
  • 6. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES SORT THESE COLORS! © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 3
  • 7. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES SORT THESE COLORS! © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 3
  • 8. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES SORT THESE COLORS! © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 4 What if there are a lot of colors to be sorted?
  • 9. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES SORT THESE COLORS! © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 4 What if there are a lot of colors to be sorted? Or, what if you have to sort two palettes at the same time?
  • 10. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 5 THIS PAPER INTRODUCES considering both hue and lightness simultaneously Single Palette Sorting 1 allowing palettes to be aligned concurrently Multiple Palettes Sorting 2 reflecting sorted color tendencies Similarity Measurement between Palettes 3 THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES
  • 11. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Lexicographical Sorting according to the Color Space Previous Sorting Methods #1 LEXICOGRAPHICAL SORTING © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 6 RGB HSV VHS LAB Pros Cons  Between lightness and hue, only one priority can be employed.  Hue is a circular quantity with discontinuity at 0 and 360 degrees.  Simple and easy-to-compute Input Palette Sort
  • 12. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Previous Sorting Methods #2 IMAGE COMPRESSION © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 7 Purpose Reflecting pixels’ spatiality to reproduce the original as much as possible while increasing compression ratio. Methods  Geometric graph traversal by reducing to the traveling salesman problem (TSP). - Greedy algorithm, Neural network, Simulated annealing, Adaptive particle swarm optimization, Genetic algorithm, etc. Cons  Visual plausibility is not taken into account.  It cannot sort multiple palettes simultaneously. Color Sorting for the Image Compression
  • 13. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Previous Sorting Methods #3 BINARY PALETTE SORTING © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 8 Cons Purpose To sort two sets of multiple palettes Methods  Similar to merge sort  (1) Partitioning two given palettes into a single palette and (2) repeatedly sorting merging in a bottom-up manner.  It assumes that each palette has a similar color distribution pattern.  Because the colors are sorted only in a row-chunk unit, it is difficult to arrange the multiple palettes naturally if they are composed of unorganized colors. Binary Palette Sorting (BPS) [Phan et al. 2018]
  • 14. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES METHODS © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 9
  • 15. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES SORTING STRATEGY • We define a natural order of colors as adjacent colors have smooth hue and lightness changes. © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 10 Image source: mbaumi @Unsplash
  • 16. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES SORTING STRATEGY • We define a natural order of colors as adjacent colors have smooth hue and lightness changes. • To reflect both conditions, we use a multi-dimensional graph traversal with colors as vertices and connecting orders as edges. © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 10 Image source: cosminserban @Unsplash
  • 17. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES SORTING STRATEGY • We define a natural order of colors as adjacent colors have smooth hue and lightness changes. • To reflect both conditions, we use a multi-dimensional graph traversal with colors as vertices and connecting orders as edges. • We adopt CIELAB color space which is more perceptually uniform than other color spaces. © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 10 Image source: Holger kkk Everding CC BY-SA @Wikipedia
  • 18. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES SORTING STRATEGY • We define a natural order of colors as adjacent colors have smooth hue and lightness changes. • To reflect both conditions, we use a multi-dimensional graph traversal with colors as vertices and connecting orders as edges. • We adopt CIELAB color space which is more perceptually uniform than other color spaces. • Graph traversal is reduced to symmetric TSP using CIEDE2000 distance. © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 10 Image source: n-voitkevich @Pexels
  • 19. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Method #1 SINGLE PALETTE SORTING © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 11
  • 20. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Method #1 SINGLE PALETTE SORTING © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 11 TSP Solver LKH
  • 21. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Method #1 SINGLE PALETTE SORTING © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 12
  • 22. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Method #2 SORTING A PAIR OF PALETTES What if you need to sort two palettes at the same time? © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 13
  • 23. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Method #2 SORTING A PAIR OF PALETTES What if you need to sort two palettes at the same time? © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 13 We arrange the color swatches to satisfy two conditions: Overall color flow is similar when the palettes are placed side by side. Each palette is naturally sorted.
  • 24. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Method #2 SORTING A PAIR OF PALETTES © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 14
  • 25. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Method #2 SORTING A PAIR OF PALETTES © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 14 TSP Solver LKH Where to Cut?
  • 26. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Method #2 SORTING A PAIR OF PALETTES © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 14 SPS Cutting Strategy TSP Solver LKH
  • 27. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Method #2 SORTING A PAIR OF PALETTES © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 15 SPS Cutting Strategy Input Palettes Sorted Palettes Merged Palettes The most different pair of consecutive colors. Radical color change cannot be avoided!
  • 28. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Method #2 SORTING A PAIR OF PALETTES © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 15 Input Palettes Cutting Strategy for a Pair of Palettes  Arranging the two palettes in a similar color flow.  Preventing sudden color changes in each palette. Starting from the point of view of individual palettes. Purpose Methods
  • 29. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Method #2 SORTING A PAIR OF PALETTES © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 15 Input Palettes Cutting Strategy for a Pair of Palettes  Arranging the two palettes in a similar color flow.  Preventing sudden color changes in each palette. Starting from the point of view of individual palettes. Purpose Methods
  • 30. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Method #2 SORTING A PAIR OF PALETTES © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 15 Input Palettes Cutting Strategy for a Pair of Palettes  Arranging the two palettes in a similar color flow.  Preventing sudden color changes in each palette. Starting from the point of view of individual palettes. Purpose Methods
  • 31. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Method #2 SORTING A PAIR OF PALETTES © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 16 Cutting Strategy for a Pair of Palettes 1. Find the longest edge connecting the same palette colors.
  • 32. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Method #2 SORTING A PAIR OF PALETTES © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 16 Cutting Strategy for a Pair of Palettes 𝒄𝒂 𝒄𝒃 2. Among the two subpaths connecting 𝑐𝑎 and 𝑐𝑏, we choose the one with fewer color vertices. 1. Find the longest edge connecting the same palette colors.
  • 33. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Method #2 SORTING A PAIR OF PALETTES © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 16 Cutting Strategy for a Pair of Palettes 𝒄𝒂 𝒄𝒃 3. To decide which side the remaining colors should be bound to, we find a point 𝑐𝑘 that minimizes where 𝑎 ≤ 𝑘 ≤ 𝑏. 1. Find the longest edge connecting the same palette colors. 2. Among the two subpaths connecting 𝑐𝑎 and 𝑐𝑏, we choose the one with fewer color vertices. 𝑖=𝑎+1 𝑘 𝑑 𝑐𝑎, 𝑐𝑖 + 𝑖=𝑘+1 𝑏−1 𝑑(𝑐𝑏, 𝑐𝑖) , 𝒄𝒌 𝒄𝒌+𝟏
  • 34. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Method #2 SORTING A PAIR OF PALETTES © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 16 Cutting Strategy for a Pair of Palettes Input Palettes Sorted Palettes Merged Palettes
  • 35. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Method #2 SORTING A PAIR OF PALETTES © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 17 Input Palettes Sorted Palettes SPS Cutting Strategy PPS Cutting Strategy
  • 36. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Method #2 SORTING A PAIR OF PALETTES © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 18 Input Palettes PPS Cutting Strategy SPS Cutting Strategy
  • 37. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Method #2 SORTING A PAIR OF PALETTES © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 19 Palettes with Different Length More than Two Palettes
  • 38. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Method #3 PALETTE SIMILARITY MEASUREMENT How much do they look similar? © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 20
  • 39. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Method #3 PALETTE SIMILARITY MEASUREMENT © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 21 Previous Similarity Measurements Element-wise Similarity
  • 40. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Method #3 PALETTE SIMILARITY MEASUREMENT © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 22 Element-wise Similarity Flow-based Similarity
  • 41. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Method #3 PALETTE SIMILARITY MEASUREMENT We proposed a Dynamic Closest Color Warping (DCCW). − Geometric simularity − Not element-wise, but flow-based measurement − Regarding a palette as a sequence, and then measure the distance between two sequences © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 23 Input Pair Sorted Pair Geometric Arrangement
  • 42. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Method #3 PALETTE SIMILARITY MEASUREMENT © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 24 Sort Input Palettes Sorted Palettes
  • 43. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Method #3 PALETTE SIMILARITY MEASUREMENT © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 25 Sorted Palettes 𝐷𝐷𝐶𝐶𝑊 𝐴, 𝐵 = 𝑊 𝐴, 𝐵 + 𝑊(𝐵, 𝐴) 𝐴 + 𝐵 𝑊 𝐴, 𝐵 = 𝑎∈𝐴 min 𝑏𝑖∈𝐵 𝑑(𝑎, 𝑏𝑖𝑏𝑖+1) A B 𝒂𝟏 𝒂𝟐 𝒂𝟑 𝒂𝟒 𝒂𝟓 𝒂𝟔 𝒂𝟕 𝒂𝟖 𝒂𝟗 𝒂𝟏𝟎 𝒃𝟏 𝒃𝟐 𝒃𝟑 𝒃𝟒 𝒃𝟓 𝒃𝟔 𝒃𝟕 𝒃𝟖 𝒃𝟗 𝒃𝟏𝟎 𝑑(𝑎, 𝑏𝑖𝑏𝑖+1) 𝒃𝒊+𝟏 𝒃𝒊 𝒂 The shortest Euclidean distance between CIELAB color 𝑎 and line segment bi𝑏𝑖+1.
  • 44. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES EVALUATION © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 26
  • 45. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES EVALUATION OVERVIEW © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 27 Single Palette Sorting SPS Palette Pair Sorting PPS Similarity Measurement DCCW  How naturally the colors are aligned  How well the colors are aligned between palettes  How naturally the colors are aligned in each palette  How effectively DCCW rates similar palettes as similar  How effectively DCCW rates dissimilar palettes as dissimilar
  • 46. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Evaluation #1 SINGLE PALETTE SORTING © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 28 Single Palette Sorting SPS  How naturally the colors are aligned If there is a correctly ordered palette, we can evaluate how well SPS aligned a shuffled palette comparing to a ground-truth palette. Ground Truth (Correctly Ordered) Sorted Palette
  • 47. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 29 Farnsworth-Munsell 100 Hue Test • It is designed to test for color blindness, with 85 different hues. • It tests the ability to distinguish and arrange minute hue differences under constant value and chroma. Dataset #1: Farnsworth-Munsell 100 Hue Test Palette Dataset (FM100P) Evaluation #1 SINGLE PALETTE SORTING
  • 48. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Evaluation #1 SINGLE PALETTE SORTING © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 30 Farnsworth-Munsell 100 Hue Test Dataset #1: Farnsworth-Munsell 100 Hue Test Palette Dataset (FM100P) Correctly Sorted Palette (Ground Truth) We generated several FM100P types with palette lengths 10, 15, 20, 25, 30, 35, and 40, each comprising 100 palettes. Maximum Distance: 5
  • 49. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 31 📐 How can we measure the accuracy of the sorting result? Ground Truth (GT) Target Success = The order of the GT and target palette is identical. Success Rate = (# of successes) / (# of tests) 1. Success Rate The longest subsequence length for the sorted elements, where the subsequences are not necessarily contiguous in the sequence 3. Length of the Longest Increasing Subsequence (LLIS) Measuring the difference between two sequences. 2. Levenshtein Distance To measure differences between two palettes even if the sorted order is not perfectly correct. Evaluation #1 SINGLE PALETTE SORTING: ACCURACY
  • 50. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Evaluation #1 SINGLE PALETTE SORTING: RESULT © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 32 CIELAB HSV GA FIA SA ACO-50 ACO-10 ACO-2 LKH (Ours) SR 0.001 0.829 0.201 0.011 0.020 0.733 0.631 0.540 0.789 LD 0.884 0.014 0.760 0.849 0.874 0.562 0.562 0.574 0.563 LLIS 0.488 0.993 0.693 0.623 0.463 0.967 0.954 0.941 0.966 ET 0.001 0.001 0.201 0.129 0.294 4.889 1.035 0.552 0.169 Since FM100P was originally sorted by hue value, it is expected that HSV lexicographical sorting performs best Time Complexity of ACO: 𝑂(𝐼n2 𝑚) Time Complexity of LKH: 𝑂(𝐼𝑛2.2 ) I: # of iterations, m: the number of ants SR: Success Rate LD: Levenshetin Distance ET: Elapsed Time
  • 51. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES EVALUATION OVERVIEW © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 33 Single Palette Sorting SPS Palette Pair Sorting PPS Similarity Measurement DCCW  How naturally the colors are aligned  How well the colors are aligned between palettes  How naturally the colors are aligned in each palette  How effectively DCCW rates similar palettes as similar  How effectively DCCW rates dissimilar palettes as dissimilar
  • 52. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Evaluation #2 PALETTE PAIR SORTING: DATASET © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 34 Palette Pair Sorting PPS  How well the colors are aligned between palettes  How naturally the colors are aligned in each palette FM100P for SPS Evaluation
  • 53. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Evaluation #2 PALETTE PAIR SORTING: DATASET © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 35 hue tone Kobayashi Hue & Tone 130 System • It is designed for color semantics based on the Munsell color system and ISCC-NBS color naming system. Neutral Dataset #2: Kobayashi Hue & Tone Palette Dataset (KHTP)
  • 54. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Evaluation #2 PALETTE PAIR SORTING: DATASET © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 36 hue tone To avoid bipartite matching, we varied row-grouped palettes by (1) mixing, (2) jittering, and (3) interpolation. Original Palette Sampling Dataset #2: Kobayashi Hue & Tone Palette Dataset (KHTP)
  • 55. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Mixing changes each palette color into a different color in the same column with a maximum adjacent index difference of 2. Evaluation #2 PALETTE PAIR SORTING: DATASET © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 37 hue tone Mixing Colors Dataset #2: Kobayashi Hue & Tone Palette Dataset (KHTP) To avoid bipartite matching, we varied row-grouped palettes by (1) mixing, (2) jittering, and (3) interpolation. Mixing
  • 56. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Evaluation #2 PALETTE PAIR SORTING: DATASET © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 38 𝛼 = 5 𝛼 = 10 𝛼 = 15 Dataset #2: Kobayashi Hue & Tone Palette Dataset (KHTP) To avoid bipartite matching, we varied row-grouped palettes by (1) mixing, (2) jittering, and (3) interpolation. Jittering Jittering adds an arbitrary offset, Ω(−𝛼, 𝛼), to each color to provide minute color changes.
  • 57. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Evaluation #2 PALETTE PAIR SORTING: DATASET © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 39 Dataset #2: Kobayashi Hue & Tone Palette Dataset (KHTP) To avoid bipartite matching, we varied row-grouped palettes by (1) mixing, (2) jittering, and (3) interpolation. Interpolation Interpolation adds additional colors to the palette to disrupt bipartite matching. #Interpolation = 1 #Interpolation = 3 #Interpolation = 5
  • 58. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Evaluation #2 PALETTE PAIR SORTING: ACCURACY © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 40 📐 How can we measure whether a pair of palettes are correctly sorted? Ground Truth Pair Sorted Pair A1 A2 A3 A4 A5 A6 A7 A8 A9 A10 A11 B1 B2 B3 B4 B5 B6 B7 B8 B9 B10 B11 A3 A7 A6 A9 A2 A5 A8 A10 A4 A11 A1 B9 B3 B5 B6 B11 B1 B7 B4 B8 B2 B10
  • 59. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Evaluation #2 PALETTE PAIR SORTING: ACCURACY © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 41  (Concurrency) How well the colors are aligned between palettes LLCS indicates the length of the longest subsequence, which is common to two sequences, and the subsequence is not necessarily consecutive. 2. Length of the longest common subsequence (LLCS) Measuring the difference between two sequences. 1. Levenshtein Distance Sorted Pair A3 A7 A6 A9 A2 A5 A8 A10 A4 A11 A1 B9 B3 B5 B6 B11 B1 B7 B4 B8 B2 B10 📐 How can we measure whether a pair of palettes are correctly sorted?
  • 60. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Evaluation #2 PALETTE PAIR SORTING: ACCURACY © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 42  (Naturalness) How naturally the colors were aligned in each palette Sorted A3 A7 A6 A9 A2 A5 A8 A10 A4 A11 A1 B9 B3 B5 B6 B11 B1 B7 B4 B8 B2 B10 Same metric with SPS 4. LLIS of Each Palette Measuring the difference between sorted and ground truth palette 3. Levenshtein Distance of Each Palette Ground Truth A1 A2 A3 A4 A5 A6 A7 A8 A9 A10 A11 B1 B2 B3 B4 B5 B6 B7 B8 B9 B10 B11 Sorted Ground Truth 📐 How can we measure whether a pair of palettes are correctly sorted?
  • 61. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Evaluation #2 PALETTE PAIR SORTING: RESULT © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 43 Ours BPS I-BPS Naturalness LD 0.631 0.783 0.651 LLIS 0.486 0.455 0.475 Concurrency LD 0.375 0.603 0.430 LLCS 0.769 0.560 0.738 Accuracy of the Palette Pair Sorting LD: Levenshetin Distance
  • 62. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Evaluation #2 PALETTE PAIR SORTING: RESULT © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 44 w/o cutting strategy w/ cutting strategy Δ𝑚𝑎𝑥 Δ𝑎𝑣𝑔 Δ𝑚𝑎𝑥 Δ𝑎𝑣𝑔 w/o jittering and interpolation 27.947 12.767 26.902 12.743 w/ jittering only 29.285 16.529 28.463 16.465 w/ interpolation only 28.438 13.076 27.561 13.071 w/ jittering and interpolation 29.015 16.346 27.901 16.257 𝛥𝑎𝑣𝑔: the average difference of adjacent colors 𝛥𝑚𝑎𝑥: the maximum color difference Effects of the Cutting Strategy
  • 63. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES EVALUATION OVERVIEW © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 45 Single Palette Sorting SPS Palette Pair Sorting PPS Similarity Measurement DCCW  How naturally the colors are aligned  How well the colors are aligned between palettes  How naturally the colors are aligned in each palette  How effectively DCCW rates similar palettes as similar  How effectively DCCW rates dissimilar palettes as dissimilar
  • 64. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Evaluation #3 DCCW: DATASET © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 46 Similarity Measurement DCCW  How effectively DCCW rates similar palettes as similar  How effectively DCCW rates dissimilar palettes as dissimilar FM100P for SPS Evaluation KHTP for PPS Evaluation
  • 65. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Evaluation #3 DCCW: DATASET © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 47 Lin and Hanrahan’s Palettes Dataset • [Lin and Hanrahan 2013] generated a color palette dataset to model human’s color palette recognition from images. • The dataset comprises unsorted palettes extracted from 40 images. • by humans including artists and Amazon Mechanical Turk’s participants • by machines using k-means, c-means clustering, and their extraction model. Dataset #3: Lin and Hanrahan Similarity Palettes (LHSP)
  • 66. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Evaluation #3 DCCW: DATASET © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 47 Lin and Hanrahan’s Palettes Dataset • [Lin and Hanrahan 2013] generated a color palette dataset to model human’s color palette recognition from images. • The dataset comprises unsorted palettes extracted from 40 images. • by humans including artists and Amazon Mechanical Turk’s participants • by machines using k-means, c-means clustering, and their extraction model. Dataset #3: Lin and Hanrahan Similarity Palettes (LHSP) Their Models Amazon MTurk Artists Source Colors Source Image
  • 67. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Evaluation #3 DCCW: DATASET © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 47 Lin and Hanrahan’s Palettes Dataset • [Lin and Hanrahan 2013] generated a color palette dataset to model human’s color palette recognition from images. • The dataset comprises unsorted palettes extracted from 40 images. • by humans including artists and Amazon Mechanical Turk’s participants • by machines using k-means, c-means clustering, and their extraction model. Dataset #3: Lin and Hanrahan Similarity Palettes (LHSP)
  • 68. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Evaluation #3 DCCW: DATASET © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 48 • To make palettes composed of similar but different colors • Replacement: replaced 10% of the colors of the palette with arbitrary colors. • Jittering: jittered the colors with jittering 𝛼 ranges = 0, 5, 10, 15 Source Image Example of Generated Palettes Dataset #3: Lin and Hanrahan Similarity Palettes (LHSP)
  • 69. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Evaluation #3 DCCW: DATASET © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 49 • We divided the palettes from the same image into disjoint query and retrieval sets. • We generated several LHSP types with palette lengths 5, 10, 15, 20, 25, and 30. • LHSP with palette lengths 10~30: • 50 query palettes (from 10 images) • 400 retrieval palettes (from 40 images including 10 query images) • LHSP with palette length 5 • 10 query palettes (from 10 images) • 1,600 retrieval palettes (from 40 images including 10 query images) Dataset #3: Lin and Hanrahan Similarity Palettes (LHSP)
  • 70. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Evaluation #3 DCCW: RESULT © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 50 📐 How well similar palettes from the same image are retrieved among the dataset. DCCW MPD HD MHD LTS-HD MCD MPHSM CEMD MBME DTW w/o jittering and replacement 0.888 0.219 0.571 0.874 0.888 0.880 0.016 0.806 0.813 0.811 w/ jittering only 0.785 0.206 0.524 0.771 0.701 0.763 0.015 0.713 0.713 0.712 w/ replacement only 0.857 0.208 0.206 0.845 0.874 0.857 0.016 0.777 0.788 0.784 w/ jittering and replacement 0.674 0.189 0.197 0.662 0.631 0.667 0.016 0.630 0.631 0.630 Mean Average Precision (mAP) for LHSP
  • 71. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Evaluation #3 DCCW: RESULT © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 51 DCCW MPD HD MHD LTS-HD MCD MPHSM CEMD MBME DTW 𝑘 = 5 0.618 0.152 0.345 0.571 0.509 0.588 0.017 0.454 0.457 0.418 𝑘 = 10 0.549 0.176 0.304 0.529 0.484 0.522 0.016 0.429 0.432 0.447 𝑘 = 15 0.683 0.189 0.293 0.678 0.654 0.671 0.015 0.626 0.629 0.632 𝑘 = 20 0.862 0.217 0.371 0.849 0.828 0.848 0.015 0.801 0.804 0.818 𝑘 = 25 0.917 0.237 0.418 0.923 0.904 0.923 0.015 0.924 0.925 0.924 𝑘 = 30 0.963 0.238 0.474 0.963 0.939 0.966 0.015 0.976 0.976 0.976 Mean Average Precision (mAP) for LHSP according to different palette size 𝑘 📐 How well similar palettes from the same image are retrieved among the dataset.
  • 72. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Evaluation #4 PERCEPTUAL STUDY © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 52 Ours  Verifying that our method is superior for palette sorting Purpose Method  Amazon Mechanical Turk Questionnaire SPS Q. Which one do you think is arranged in the most natural order? HSV-based Sorting CIELAB-based Sorting
  • 73. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Evaluation #4 PERCEPTUAL STUDY © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 53 Ours  Verifying that our method is superior for palette sorting Purpose Method  Amazon Mechanical Turk Questionnaire PPS Q. Which pair do you think is arranged in the most natural order? HSV-based Sorting CIELAB-based Sorting BPS I-BPS
  • 74. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Evaluation #4 PERCEPTUAL STUDY: RESULT © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 54 𝑘 = 10 𝑘 = 15 𝑘 = 20 SPS Ours > HSV-based sorting 79.9% 85.9% 86.7% Ours > CIELAB-based sorting 69.3% 63.9% 64.4% PPS Ours > HSV-based sorting 82.9% 88.8% 92.1% Ours > CIELAB-based sorting 64.1% 60.9% 59.8% Ours > BPS 86.5% 89.0% 91.4% Ours > I-BPS 67.4% 64.9% 71.0% * The effects of palette length and sorting methods were determined using a two-way repeated measures ANOVA with a significance level of 𝑝 < 0.001 for both SPS and PPS results
  • 75. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES APPLICATIONS © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 55
  • 76. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Application #1 PALETTE INTERPOLATION © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 56 HSV-based CIELAB-based DCCW-based
  • 77. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Application #2 PALETTE NAVIGATION © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 57 Kiki's Delivery Service (1989) Swallow (2019) Call Me by Your Name (2017) The Place Beyond the Pines (2012) Pan's Labyrinth (2006) Dunkirk (2017) Query Palette Unsplash WikiArt COLOURlovers Commercial Movie
  • 78. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Django Unchained (2012) Swallow (2019) The Master (2012) The Neon Demon (2016) Titanic (1997) World War Z (2013) Application #2 PALETTE NAVIGATION © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 58 Query Palette Unsplash WikiArt COLOURlovers Commercial Movie
  • 79. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Application #3 PALETTE-BASED RECOLORING © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 59 Recoloring Convex Hull Deformation Original Image Extracted Palette Convex Hull Target Palette Simple Transformation Extending [Tan et al. 2018]
  • 80. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Application #3 PALETTE-BASED RECOLORING © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 60 Original Image Target Palette Improved Transformation Recoloring Result Convex Hull Deformation Palette Pair Sorting Source (Extracted) Target 1 Sorted Source Sorted target 2 Nearest Matching Using DCCW Closest Colors from Target to Source 3 Sorted Target 4 Buffered Palette Merging
  • 81. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES Application #3 PALETTE-BASED RECOLORING © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 61 Target Palette Simple Transformation Improved Transformation Original Image Target Palette Simple Transformation Improved Transformation Original Image
  • 82. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES CONCLUSION © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 62
  • 83. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES CONCLUSION • We proposed a method to sort palette colors by traversing color graphs. • We proposed a dynamic closest color warping (DCCW) similarity measurement for palette comparison. • We evaluated our palette sorting and DCCW on various datasets using various criteria confirming ours outperformed previous methods in terms of accuracy and computation time. • We also evaluated our methods using a perceptual study, and the result clearly indicates that ours generate perceptually natural sorting results, outperforming previous techniques. • We demonstrate many compelling applications enabled by DCCW, including palette interpolation, palette navigation, and image recoloring. • Source code and dataset are available at: • https://github.com/SuziKim/DCCW / https://github.com/SuziKim/DCCW-dataset © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 63
  • 84. THE PREMIER CONFERENCE & EXHIBITION IN COMPUTER GRAPHICS & INTERACTIVE TECHNIQUES © 2021 SIGGRAPH. ALL RIGHTS RESERVED. 64 THANK YOU! SUZI KIM AND SUNGHEE CHOI GEOMETRIC COMPUTING LAB. SCHOOL OF COMPUTING, KAIST DYNAMIC CLOSEST COLOR WARPING TO SORT AND COMPARE PALETTES