SlideShare a Scribd company logo
Deep learning for
satellite imagery
colorization and
distance measuring
Lviv AI&BigData Day, November 4
Volodymyr Getmanskyi
ELEKS data science team
skype: paradoxx_xx
Automatic image colorization is the task of adding colors to a new grayscale image without any
user intervention. This problem is ill-posed in the sense that there is no unique colorization of a
grayscale image without any prior knowledge. Indeed, many objects can have different colors:
artificial objects, such as plastic objects which can have random colors, but also for natural objects
such as tree leaves which can have various nuances of green and brown in different seasons,
without significant change of shape.
Most of the modern methods uses pretrained convolutional
neural networks and extracts the information about
features/objects from the image. Then there is a possibility to
upscale them to original size and concatenating them all
together.
Also most of them uses some simplification to train the models to
produce two color channels which can be concatenated with the
grayscale input channel to produce the YUV/CIELUV image.
The most obvious loss function is a Euclidean distance function between the network RGB output image
and the true color RGB image (distance in UV space – from previous slide).
We, however, notice that in numerous cases, the images colorized with such approach are mostly sepia-
toned and muted in color.
To understand why, consider a pixel that exists in a flower petal across multiple images that are identical,
save for the color of the flower petals. Depending on the picture, this pixel can take on various tones of
red, yellow, blue, and more. With a regression-based system that uses an loss function, the predicted
pixel value that minimizes the loss for this particular pixel is the mean pixel value. Accordingly, the
predicted pixel ends up being a mixture of the possible colors.
Another problem is that pre-trained CNN weights, that was used for feature extraction, are from
estimators that were trained on reсognizing different objects but not the objects from satellite images.
Also all trained models for colorizing (with RGB/UV distance minimizing), were trained to restore
colors for different objects but not for satellite images.
We tried to test existing solutions but most of them are facing the problems that we mentioned earlier
and demonstrates low performance:
* http://demos.algorithmia.com/colorize-photos, http://www.colorizephoto.com/converter
http://pinetools.com/colorize-image, https://github.com/richzhang/colorization
After the first steps we found more efficient ways (with no or small sepia effects) and also try to train
the estimator based on maps.
Below you can see our basic results with adding color histograms and color mapping as additional map
features:
Below you can see our basic results with models training (black&white original, colorized, original):
First iteration of our model (sepia warning): Weak trained model (1K iteration, 20K
samples without augmentation):
Measuring distance is a key tool in map reading and is especially useful for hikers and cyclists who want
to measure how far they have travelled or how far they wish to go based on raw maps. BTW, such
distance measuring is helpful for understanding map scale that is the first step for camera height
evaluation (the distance from sea level to eye level) or altitude of the plane/drone above sea / terrain level
(ASL).
So the main task here is to get the scale. If you understand the map scale number that's great and you
can still measure distances accurately.
So based on such thoughts we decided to build the estimator that can measure distances on real
unlabeled maps using some patterns like shadows, trees’ crowns, roofs and roads.
For data gathering we used typical browser and google maps with ImageGrab and win32api python
libraries the same as for colorization task (20K samples), but there are few image processing methods
before the DL usage. So here we need to extract/recognize the scale that will be our labels. Scale is
located on the right bottom of the image, so we need to detect the defining point from which we can move
scale detection. Here we can simply slides the template image over the input image (as in cascade
methods) and compares the template and patch of input image with some distance metric.
After the successful detection we can extract two elements that we need for labeling:
- Current measurement units (using simple OCR like tesseract and further rule-based engine)
- Scale Line length (simply calculated the location of rows’ mode with continuous range of white
columns)
After the validation comparison with VGG architecture we choose AlexNet architecture with all
convolutional layers frozen (transfer learning for efficient feature extraction):
# Real-time data preprocessing
img_prep = ImagePreprocessing()
img_prep.add_featurewise_zero_center()
img_prep.add_featurewise_stdnorm()
# Real-time data augmentation
img_aug = ImageAugmentation()
img_aug.add_random_flip_leftright()
img_aug.add_random_rotation(max_angle=135.)
Results (examples from testset, mae≈25%):
Image Side length Predicted length Image Side length Predicted length
168 m 213 m 45 m 69 m
1333 m 954 m 4970 m 4720 m
Through our small experiment, we have checked the efficiency of using deep neural networks to colorize
black and white satellite images the same as for measuring maps’ scale. So our conclusions are the
following.
In particular, formulating the maps colorization task as a classification problem can yield colorized maps
that are arguably much more aesthetically-pleasing than those generated by a baseline regression-based
model (with sepia), and thus shows much promise for further development.
Also, redesigning the system around an adversarial network may improve results, since instead of
focusing on minimizing the loss on a per-pixel deviation, the system would learn to generate/colorize
pictures that compare well with real-world maps.
From distance measuring approach we recommend to use similar feature extraction (transfer learning)
that helps to feed the real objects (trees, roofs, roads) into distance result.
Inspired by Technology.
Driven by Value.
Find us at eleks.com
Have a question? Write to eleksinfo@eleks.com

More Related Content

What's hot

Segmentation Techniques -I
Segmentation Techniques -ISegmentation Techniques -I
Segmentation Techniques -I
Hemantha Kulathilake
 
3D Reconstruction from Multiple uncalibrated 2D Images of an Object
3D Reconstruction from Multiple uncalibrated 2D Images of an Object3D Reconstruction from Multiple uncalibrated 2D Images of an Object
3D Reconstruction from Multiple uncalibrated 2D Images of an Object
Ankur Tyagi
 
Image Segmentation from RGBD Images by 3D Point Cloud Attributes and High-Lev...
Image Segmentation from RGBD Images by 3D Point Cloud Attributes and High-Lev...Image Segmentation from RGBD Images by 3D Point Cloud Attributes and High-Lev...
Image Segmentation from RGBD Images by 3D Point Cloud Attributes and High-Lev...
CSCJournals
 
6 texture mapping computer graphics
6 texture mapping computer graphics6 texture mapping computer graphics
6 texture mapping computer graphics
cairo university
 
Digital image processing
Digital image processingDigital image processing
Digital image processing
Deevena Dayaal
 
V2 v posenet
V2 v posenetV2 v posenet
V2 v posenet
NAVER Engineering
 
Seed net automatic seed generation with deep reinforcement learning for robus...
Seed net automatic seed generation with deep reinforcement learning for robus...Seed net automatic seed generation with deep reinforcement learning for robus...
Seed net automatic seed generation with deep reinforcement learning for robus...
NAVER Engineering
 
Intel, Intelligent Systems Lab: Syable View Synthesis Whitepaper
Intel, Intelligent Systems Lab: Syable View Synthesis WhitepaperIntel, Intelligent Systems Lab: Syable View Synthesis Whitepaper
Intel, Intelligent Systems Lab: Syable View Synthesis Whitepaper
Alejandro Franceschi
 
Computer Graphics Introduction
Computer Graphics IntroductionComputer Graphics Introduction
Computer Graphics IntroductionGhaffar Khan
 
Digital Image Processing
Digital Image ProcessingDigital Image Processing
Digital Image Processing
Sahil Biswas
 
An Open Source solution for Three-Dimensional documentation: archaeological a...
An Open Source solution for Three-Dimensional documentation: archaeological a...An Open Source solution for Three-Dimensional documentation: archaeological a...
An Open Source solution for Three-Dimensional documentation: archaeological a...
Giulio Bigliardi
 
Object Based Image Analysis
Object Based Image Analysis Object Based Image Analysis
Object Based Image Analysis
Kabir Uddin
 
Video processing on dsp
Video processing on dspVideo processing on dsp
Video processing on dsp
Nirma University
 
High Performance Computing for Satellite Image Processing and Analyzing – A ...
High Performance Computing for Satellite Image  Processing and Analyzing – A ...High Performance Computing for Satellite Image  Processing and Analyzing – A ...
High Performance Computing for Satellite Image Processing and Analyzing – A ...
Editor IJCATR
 
CPlaNet: Enhancing Image Geolocalization by Combinatorial Partitioning of Maps
CPlaNet: Enhancing Image Geolocalization by Combinatorial Partitioning of MapsCPlaNet: Enhancing Image Geolocalization by Combinatorial Partitioning of Maps
CPlaNet: Enhancing Image Geolocalization by Combinatorial Partitioning of Maps
NAVER Engineering
 
Passive stereo vision with deep learning
Passive stereo vision with deep learningPassive stereo vision with deep learning
Passive stereo vision with deep learning
Yu Huang
 
JPM1407 Exposing Digital Image Forgeries by Illumination Color Classification
JPM1407   Exposing Digital Image Forgeries by Illumination Color ClassificationJPM1407   Exposing Digital Image Forgeries by Illumination Color Classification
JPM1407 Exposing Digital Image Forgeries by Illumination Color Classification
chennaijp
 
Digital image processing
Digital image processingDigital image processing
Digital image processing
lakhveer singh
 
Image processing
Image processingImage processing
Image processing
Raga Deepthi
 

What's hot (20)

Assignment 2
Assignment 2Assignment 2
Assignment 2
 
Segmentation Techniques -I
Segmentation Techniques -ISegmentation Techniques -I
Segmentation Techniques -I
 
3D Reconstruction from Multiple uncalibrated 2D Images of an Object
3D Reconstruction from Multiple uncalibrated 2D Images of an Object3D Reconstruction from Multiple uncalibrated 2D Images of an Object
3D Reconstruction from Multiple uncalibrated 2D Images of an Object
 
Image Segmentation from RGBD Images by 3D Point Cloud Attributes and High-Lev...
Image Segmentation from RGBD Images by 3D Point Cloud Attributes and High-Lev...Image Segmentation from RGBD Images by 3D Point Cloud Attributes and High-Lev...
Image Segmentation from RGBD Images by 3D Point Cloud Attributes and High-Lev...
 
6 texture mapping computer graphics
6 texture mapping computer graphics6 texture mapping computer graphics
6 texture mapping computer graphics
 
Digital image processing
Digital image processingDigital image processing
Digital image processing
 
V2 v posenet
V2 v posenetV2 v posenet
V2 v posenet
 
Seed net automatic seed generation with deep reinforcement learning for robus...
Seed net automatic seed generation with deep reinforcement learning for robus...Seed net automatic seed generation with deep reinforcement learning for robus...
Seed net automatic seed generation with deep reinforcement learning for robus...
 
Intel, Intelligent Systems Lab: Syable View Synthesis Whitepaper
Intel, Intelligent Systems Lab: Syable View Synthesis WhitepaperIntel, Intelligent Systems Lab: Syable View Synthesis Whitepaper
Intel, Intelligent Systems Lab: Syable View Synthesis Whitepaper
 
Computer Graphics Introduction
Computer Graphics IntroductionComputer Graphics Introduction
Computer Graphics Introduction
 
Digital Image Processing
Digital Image ProcessingDigital Image Processing
Digital Image Processing
 
An Open Source solution for Three-Dimensional documentation: archaeological a...
An Open Source solution for Three-Dimensional documentation: archaeological a...An Open Source solution for Three-Dimensional documentation: archaeological a...
An Open Source solution for Three-Dimensional documentation: archaeological a...
 
Object Based Image Analysis
Object Based Image Analysis Object Based Image Analysis
Object Based Image Analysis
 
Video processing on dsp
Video processing on dspVideo processing on dsp
Video processing on dsp
 
High Performance Computing for Satellite Image Processing and Analyzing – A ...
High Performance Computing for Satellite Image  Processing and Analyzing – A ...High Performance Computing for Satellite Image  Processing and Analyzing – A ...
High Performance Computing for Satellite Image Processing and Analyzing – A ...
 
CPlaNet: Enhancing Image Geolocalization by Combinatorial Partitioning of Maps
CPlaNet: Enhancing Image Geolocalization by Combinatorial Partitioning of MapsCPlaNet: Enhancing Image Geolocalization by Combinatorial Partitioning of Maps
CPlaNet: Enhancing Image Geolocalization by Combinatorial Partitioning of Maps
 
Passive stereo vision with deep learning
Passive stereo vision with deep learningPassive stereo vision with deep learning
Passive stereo vision with deep learning
 
JPM1407 Exposing Digital Image Forgeries by Illumination Color Classification
JPM1407   Exposing Digital Image Forgeries by Illumination Color ClassificationJPM1407   Exposing Digital Image Forgeries by Illumination Color Classification
JPM1407 Exposing Digital Image Forgeries by Illumination Color Classification
 
Digital image processing
Digital image processingDigital image processing
Digital image processing
 
Image processing
Image processingImage processing
Image processing
 

Viewers also liked

Ai&bigdataconference oleksandr saienko machine learning use cases in telecom
Ai&bigdataconference oleksandr saienko machine learning use cases in telecomAi&bigdataconference oleksandr saienko machine learning use cases in telecom
Ai&bigdataconference oleksandr saienko machine learning use cases in telecom
Olga Zinkevych
 
Ai big dataconference_taras firman how to build advanced prediction with addi...
Ai big dataconference_taras firman how to build advanced prediction with addi...Ai big dataconference_taras firman how to build advanced prediction with addi...
Ai big dataconference_taras firman how to build advanced prediction with addi...
Olga Zinkevych
 
Ai big dataconference_krakovetskyi_microsoft ai a new era of smart solutions
Ai big dataconference_krakovetskyi_microsoft ai a new era of smart solutionsAi big dataconference_krakovetskyi_microsoft ai a new era of smart solutions
Ai big dataconference_krakovetskyi_microsoft ai a new era of smart solutions
Olga Zinkevych
 
Ai big dataconference_sparkinonehour_vitalii bashun
Ai big dataconference_sparkinonehour_vitalii bashunAi big dataconference_sparkinonehour_vitalii bashun
Ai big dataconference_sparkinonehour_vitalii bashun
Olga Zinkevych
 
Ai big dataconference_ml_fastdata_vitalii bondarenko
Ai big dataconference_ml_fastdata_vitalii bondarenkoAi big dataconference_ml_fastdata_vitalii bondarenko
Ai big dataconference_ml_fastdata_vitalii bondarenko
Olga Zinkevych
 
Ai big dataconference_eugene_polonichko_azure data lake
Ai big dataconference_eugene_polonichko_azure data lake Ai big dataconference_eugene_polonichko_azure data lake
Ai big dataconference_eugene_polonichko_azure data lake
Olga Zinkevych
 

Viewers also liked (6)

Ai&bigdataconference oleksandr saienko machine learning use cases in telecom
Ai&bigdataconference oleksandr saienko machine learning use cases in telecomAi&bigdataconference oleksandr saienko machine learning use cases in telecom
Ai&bigdataconference oleksandr saienko machine learning use cases in telecom
 
Ai big dataconference_taras firman how to build advanced prediction with addi...
Ai big dataconference_taras firman how to build advanced prediction with addi...Ai big dataconference_taras firman how to build advanced prediction with addi...
Ai big dataconference_taras firman how to build advanced prediction with addi...
 
Ai big dataconference_krakovetskyi_microsoft ai a new era of smart solutions
Ai big dataconference_krakovetskyi_microsoft ai a new era of smart solutionsAi big dataconference_krakovetskyi_microsoft ai a new era of smart solutions
Ai big dataconference_krakovetskyi_microsoft ai a new era of smart solutions
 
Ai big dataconference_sparkinonehour_vitalii bashun
Ai big dataconference_sparkinonehour_vitalii bashunAi big dataconference_sparkinonehour_vitalii bashun
Ai big dataconference_sparkinonehour_vitalii bashun
 
Ai big dataconference_ml_fastdata_vitalii bondarenko
Ai big dataconference_ml_fastdata_vitalii bondarenkoAi big dataconference_ml_fastdata_vitalii bondarenko
Ai big dataconference_ml_fastdata_vitalii bondarenko
 
Ai big dataconference_eugene_polonichko_azure data lake
Ai big dataconference_eugene_polonichko_azure data lake Ai big dataconference_eugene_polonichko_azure data lake
Ai big dataconference_eugene_polonichko_azure data lake
 

Similar to Ai big dataconference_volodymyr getmanskyi colorization distance measuring

Implementing Neural Style Transfer
Implementing Neural Style Transfer Implementing Neural Style Transfer
Implementing Neural Style Transfer
Tahsin Mayeesha
 
International Journal of Computational Engineering Research(IJCER)
International Journal of Computational Engineering Research(IJCER)International Journal of Computational Engineering Research(IJCER)
International Journal of Computational Engineering Research(IJCER)
ijceronline
 
A PROJECT REPORT ON REMOVAL OF UNNECESSARY OBJECTS FROM PHOTOS USING MASKING
A PROJECT REPORT ON REMOVAL OF UNNECESSARY OBJECTS FROM PHOTOS USING MASKINGA PROJECT REPORT ON REMOVAL OF UNNECESSARY OBJECTS FROM PHOTOS USING MASKING
A PROJECT REPORT ON REMOVAL OF UNNECESSARY OBJECTS FROM PHOTOS USING MASKING
IRJET Journal
 
Simple Pendulum Experiment and Automatic Survey Grading using Computer Vision
Simple Pendulum Experiment and Automatic Survey Grading using Computer VisionSimple Pendulum Experiment and Automatic Survey Grading using Computer Vision
Simple Pendulum Experiment and Automatic Survey Grading using Computer VisionAnish Patel
 
Conception_et_realisation_dun_site_Web_d.pdf
Conception_et_realisation_dun_site_Web_d.pdfConception_et_realisation_dun_site_Web_d.pdf
Conception_et_realisation_dun_site_Web_d.pdf
SofianeHassine2
 
IRJET- Low Light Image Enhancement using Convolutional Neural Network
IRJET-  	  Low Light Image Enhancement using Convolutional Neural NetworkIRJET-  	  Low Light Image Enhancement using Convolutional Neural Network
IRJET- Low Light Image Enhancement using Convolutional Neural Network
IRJET Journal
 
Enhanced Optimization of Edge Detection for High Resolution Images Using Veri...
Enhanced Optimization of Edge Detection for High Resolution Images Using Veri...Enhanced Optimization of Edge Detection for High Resolution Images Using Veri...
Enhanced Optimization of Edge Detection for High Resolution Images Using Veri...
ijcisjournal
 
Automatic Building detection for satellite Images using IGV and DSM
Automatic Building detection for satellite Images using IGV and DSMAutomatic Building detection for satellite Images using IGV and DSM
Automatic Building detection for satellite Images using IGV and DSM
Amit Raikar
 
Road signs detection using voila jone's algorithm with the help of opencv
Road signs detection using voila jone's algorithm with the help of opencvRoad signs detection using voila jone's algorithm with the help of opencv
Road signs detection using voila jone's algorithm with the help of opencv
MohdSalim34
 
Comparative Study and Analysis of Image Inpainting Techniques
Comparative Study and Analysis of Image Inpainting TechniquesComparative Study and Analysis of Image Inpainting Techniques
Comparative Study and Analysis of Image Inpainting Techniques
IOSR Journals
 
Image segmentation using advanced fuzzy c-mean algorithm [FYP @ IITR, obtaine...
Image segmentation using advanced fuzzy c-mean algorithm [FYP @ IITR, obtaine...Image segmentation using advanced fuzzy c-mean algorithm [FYP @ IITR, obtaine...
Image segmentation using advanced fuzzy c-mean algorithm [FYP @ IITR, obtaine...
Koteswar Rao Jerripothula
 
Traffic Automation System
Traffic Automation SystemTraffic Automation System
Traffic Automation System
Prabal Chauhan
 
An effective RGB color selection for complex 3D object structure in scene gra...
An effective RGB color selection for complex 3D object structure in scene gra...An effective RGB color selection for complex 3D object structure in scene gra...
An effective RGB color selection for complex 3D object structure in scene gra...
IJECEIAES
 
[IJET-V1I6P16] Authors : Indraja Mali , Saumya Saxena ,Padmaja Desai , Ajay G...
[IJET-V1I6P16] Authors : Indraja Mali , Saumya Saxena ,Padmaja Desai , Ajay G...[IJET-V1I6P16] Authors : Indraja Mali , Saumya Saxena ,Padmaja Desai , Ajay G...
[IJET-V1I6P16] Authors : Indraja Mali , Saumya Saxena ,Padmaja Desai , Ajay G...
IJET - International Journal of Engineering and Techniques
 
Extraction of Buildings from Satellite Images
Extraction of Buildings from Satellite ImagesExtraction of Buildings from Satellite Images
Extraction of Buildings from Satellite Images
Akanksha Prasad
 
I010634450
I010634450I010634450
I010634450
IOSR Journals
 
Performance of Efficient Closed-Form Solution to Comprehensive Frontier Exposure
Performance of Efficient Closed-Form Solution to Comprehensive Frontier ExposurePerformance of Efficient Closed-Form Solution to Comprehensive Frontier Exposure
Performance of Efficient Closed-Form Solution to Comprehensive Frontier Exposure
iosrjce
 
Blind Source Camera Identification
Blind Source Camera Identification Blind Source Camera Identification
Blind Source Camera Identification
Sudhanshu Patel
 
Implementation of Object Tracking for Real Time Video
Implementation of Object Tracking for Real Time VideoImplementation of Object Tracking for Real Time Video
Implementation of Object Tracking for Real Time Video
IDES Editor
 

Similar to Ai big dataconference_volodymyr getmanskyi colorization distance measuring (20)

Implementing Neural Style Transfer
Implementing Neural Style Transfer Implementing Neural Style Transfer
Implementing Neural Style Transfer
 
International Journal of Computational Engineering Research(IJCER)
International Journal of Computational Engineering Research(IJCER)International Journal of Computational Engineering Research(IJCER)
International Journal of Computational Engineering Research(IJCER)
 
A PROJECT REPORT ON REMOVAL OF UNNECESSARY OBJECTS FROM PHOTOS USING MASKING
A PROJECT REPORT ON REMOVAL OF UNNECESSARY OBJECTS FROM PHOTOS USING MASKINGA PROJECT REPORT ON REMOVAL OF UNNECESSARY OBJECTS FROM PHOTOS USING MASKING
A PROJECT REPORT ON REMOVAL OF UNNECESSARY OBJECTS FROM PHOTOS USING MASKING
 
Simple Pendulum Experiment and Automatic Survey Grading using Computer Vision
Simple Pendulum Experiment and Automatic Survey Grading using Computer VisionSimple Pendulum Experiment and Automatic Survey Grading using Computer Vision
Simple Pendulum Experiment and Automatic Survey Grading using Computer Vision
 
Conception_et_realisation_dun_site_Web_d.pdf
Conception_et_realisation_dun_site_Web_d.pdfConception_et_realisation_dun_site_Web_d.pdf
Conception_et_realisation_dun_site_Web_d.pdf
 
IRJET- Low Light Image Enhancement using Convolutional Neural Network
IRJET-  	  Low Light Image Enhancement using Convolutional Neural NetworkIRJET-  	  Low Light Image Enhancement using Convolutional Neural Network
IRJET- Low Light Image Enhancement using Convolutional Neural Network
 
Enhanced Optimization of Edge Detection for High Resolution Images Using Veri...
Enhanced Optimization of Edge Detection for High Resolution Images Using Veri...Enhanced Optimization of Edge Detection for High Resolution Images Using Veri...
Enhanced Optimization of Edge Detection for High Resolution Images Using Veri...
 
Automatic Building detection for satellite Images using IGV and DSM
Automatic Building detection for satellite Images using IGV and DSMAutomatic Building detection for satellite Images using IGV and DSM
Automatic Building detection for satellite Images using IGV and DSM
 
Road signs detection using voila jone's algorithm with the help of opencv
Road signs detection using voila jone's algorithm with the help of opencvRoad signs detection using voila jone's algorithm with the help of opencv
Road signs detection using voila jone's algorithm with the help of opencv
 
Comparative Study and Analysis of Image Inpainting Techniques
Comparative Study and Analysis of Image Inpainting TechniquesComparative Study and Analysis of Image Inpainting Techniques
Comparative Study and Analysis of Image Inpainting Techniques
 
Image segmentation using advanced fuzzy c-mean algorithm [FYP @ IITR, obtaine...
Image segmentation using advanced fuzzy c-mean algorithm [FYP @ IITR, obtaine...Image segmentation using advanced fuzzy c-mean algorithm [FYP @ IITR, obtaine...
Image segmentation using advanced fuzzy c-mean algorithm [FYP @ IITR, obtaine...
 
Traffic Automation System
Traffic Automation SystemTraffic Automation System
Traffic Automation System
 
Gc2005vk
Gc2005vkGc2005vk
Gc2005vk
 
An effective RGB color selection for complex 3D object structure in scene gra...
An effective RGB color selection for complex 3D object structure in scene gra...An effective RGB color selection for complex 3D object structure in scene gra...
An effective RGB color selection for complex 3D object structure in scene gra...
 
[IJET-V1I6P16] Authors : Indraja Mali , Saumya Saxena ,Padmaja Desai , Ajay G...
[IJET-V1I6P16] Authors : Indraja Mali , Saumya Saxena ,Padmaja Desai , Ajay G...[IJET-V1I6P16] Authors : Indraja Mali , Saumya Saxena ,Padmaja Desai , Ajay G...
[IJET-V1I6P16] Authors : Indraja Mali , Saumya Saxena ,Padmaja Desai , Ajay G...
 
Extraction of Buildings from Satellite Images
Extraction of Buildings from Satellite ImagesExtraction of Buildings from Satellite Images
Extraction of Buildings from Satellite Images
 
I010634450
I010634450I010634450
I010634450
 
Performance of Efficient Closed-Form Solution to Comprehensive Frontier Exposure
Performance of Efficient Closed-Form Solution to Comprehensive Frontier ExposurePerformance of Efficient Closed-Form Solution to Comprehensive Frontier Exposure
Performance of Efficient Closed-Form Solution to Comprehensive Frontier Exposure
 
Blind Source Camera Identification
Blind Source Camera Identification Blind Source Camera Identification
Blind Source Camera Identification
 
Implementation of Object Tracking for Real Time Video
Implementation of Object Tracking for Real Time VideoImplementation of Object Tracking for Real Time Video
Implementation of Object Tracking for Real Time Video
 

More from Olga Zinkevych

Overview of text classification approaches algorithms & software v lyubin...
Overview of text classification approaches algorithms & software v lyubin...Overview of text classification approaches algorithms & software v lyubin...
Overview of text classification approaches algorithms & software v lyubin...
Olga Zinkevych
 
Evolution of words through time a malenko dataconf 21 04_18
Evolution of words through time a malenko dataconf 21 04_18Evolution of words through time a malenko dataconf 21 04_18
Evolution of words through time a malenko dataconf 21 04_18
Olga Zinkevych
 
What it takes to build a model for detecting patients that defaults from medi...
What it takes to build a model for detecting patients that defaults from medi...What it takes to build a model for detecting patients that defaults from medi...
What it takes to build a model for detecting patients that defaults from medi...
Olga Zinkevych
 
Variational autoencoders for speech processing d.bielievtsov dataconf 21 04 18
Variational autoencoders for speech processing d.bielievtsov dataconf 21 04 18Variational autoencoders for speech processing d.bielievtsov dataconf 21 04 18
Variational autoencoders for speech processing d.bielievtsov dataconf 21 04 18
Olga Zinkevych
 
Dataservices based on mesos and kafka kostiantyn bokhan dataconf 21 04 18
Dataservices based on mesos and kafka kostiantyn bokhan dataconf 21 04 18Dataservices based on mesos and kafka kostiantyn bokhan dataconf 21 04 18
Dataservices based on mesos and kafka kostiantyn bokhan dataconf 21 04 18
Olga Zinkevych
 
Azure data catalog your data your way eugene polonichko dataconf 21 04 18
Azure data catalog your data your way eugene polonichko dataconf 21 04 18Azure data catalog your data your way eugene polonichko dataconf 21 04 18
Azure data catalog your data your way eugene polonichko dataconf 21 04 18
Olga Zinkevych
 
Aibdconference chat bot for every product Maksym Volchenko
Aibdconference chat bot for every product Maksym VolchenkoAibdconference chat bot for every product Maksym Volchenko
Aibdconference chat bot for every product Maksym Volchenko
Olga Zinkevych
 
Ai big dataconference_semantic image segmentatation using word embeddings_ole...
Ai big dataconference_semantic image segmentatation using word embeddings_ole...Ai big dataconference_semantic image segmentatation using word embeddings_ole...
Ai big dataconference_semantic image segmentatation using word embeddings_ole...
Olga Zinkevych
 
Ai big dataconference_jeffrey ricker_kappa_architecture
Ai big dataconference_jeffrey ricker_kappa_architectureAi big dataconference_jeffrey ricker_kappa_architecture
Ai big dataconference_jeffrey ricker_kappa_architecture
Olga Zinkevych
 

More from Olga Zinkevych (9)

Overview of text classification approaches algorithms & software v lyubin...
Overview of text classification approaches algorithms & software v lyubin...Overview of text classification approaches algorithms & software v lyubin...
Overview of text classification approaches algorithms & software v lyubin...
 
Evolution of words through time a malenko dataconf 21 04_18
Evolution of words through time a malenko dataconf 21 04_18Evolution of words through time a malenko dataconf 21 04_18
Evolution of words through time a malenko dataconf 21 04_18
 
What it takes to build a model for detecting patients that defaults from medi...
What it takes to build a model for detecting patients that defaults from medi...What it takes to build a model for detecting patients that defaults from medi...
What it takes to build a model for detecting patients that defaults from medi...
 
Variational autoencoders for speech processing d.bielievtsov dataconf 21 04 18
Variational autoencoders for speech processing d.bielievtsov dataconf 21 04 18Variational autoencoders for speech processing d.bielievtsov dataconf 21 04 18
Variational autoencoders for speech processing d.bielievtsov dataconf 21 04 18
 
Dataservices based on mesos and kafka kostiantyn bokhan dataconf 21 04 18
Dataservices based on mesos and kafka kostiantyn bokhan dataconf 21 04 18Dataservices based on mesos and kafka kostiantyn bokhan dataconf 21 04 18
Dataservices based on mesos and kafka kostiantyn bokhan dataconf 21 04 18
 
Azure data catalog your data your way eugene polonichko dataconf 21 04 18
Azure data catalog your data your way eugene polonichko dataconf 21 04 18Azure data catalog your data your way eugene polonichko dataconf 21 04 18
Azure data catalog your data your way eugene polonichko dataconf 21 04 18
 
Aibdconference chat bot for every product Maksym Volchenko
Aibdconference chat bot for every product Maksym VolchenkoAibdconference chat bot for every product Maksym Volchenko
Aibdconference chat bot for every product Maksym Volchenko
 
Ai big dataconference_semantic image segmentatation using word embeddings_ole...
Ai big dataconference_semantic image segmentatation using word embeddings_ole...Ai big dataconference_semantic image segmentatation using word embeddings_ole...
Ai big dataconference_semantic image segmentatation using word embeddings_ole...
 
Ai big dataconference_jeffrey ricker_kappa_architecture
Ai big dataconference_jeffrey ricker_kappa_architectureAi big dataconference_jeffrey ricker_kappa_architecture
Ai big dataconference_jeffrey ricker_kappa_architecture
 

Recently uploaded

Tabula.io Cheatsheet: automate your data workflows
Tabula.io Cheatsheet: automate your data workflowsTabula.io Cheatsheet: automate your data workflows
Tabula.io Cheatsheet: automate your data workflows
alex933524
 
Levelwise PageRank with Loop-Based Dead End Handling Strategy : SHORT REPORT ...
Levelwise PageRank with Loop-Based Dead End Handling Strategy : SHORT REPORT ...Levelwise PageRank with Loop-Based Dead End Handling Strategy : SHORT REPORT ...
Levelwise PageRank with Loop-Based Dead End Handling Strategy : SHORT REPORT ...
Subhajit Sahu
 
FP Growth Algorithm and its Applications
FP Growth Algorithm and its ApplicationsFP Growth Algorithm and its Applications
FP Growth Algorithm and its Applications
MaleehaSheikh2
 
Jpolillo Amazon PPC - Bid Optimization Sample
Jpolillo Amazon PPC - Bid Optimization SampleJpolillo Amazon PPC - Bid Optimization Sample
Jpolillo Amazon PPC - Bid Optimization Sample
James Polillo
 
Adjusting primitives for graph : SHORT REPORT / NOTES
Adjusting primitives for graph : SHORT REPORT / NOTESAdjusting primitives for graph : SHORT REPORT / NOTES
Adjusting primitives for graph : SHORT REPORT / NOTES
Subhajit Sahu
 
Predicting Product Ad Campaign Performance: A Data Analysis Project Presentation
Predicting Product Ad Campaign Performance: A Data Analysis Project PresentationPredicting Product Ad Campaign Performance: A Data Analysis Project Presentation
Predicting Product Ad Campaign Performance: A Data Analysis Project Presentation
Boston Institute of Analytics
 
standardisation of garbhpala offhgfffghh
standardisation of garbhpala offhgfffghhstandardisation of garbhpala offhgfffghh
standardisation of garbhpala offhgfffghh
ArpitMalhotra16
 
SOCRadar Germany 2024 Threat Landscape Report
SOCRadar Germany 2024 Threat Landscape ReportSOCRadar Germany 2024 Threat Landscape Report
SOCRadar Germany 2024 Threat Landscape Report
SOCRadar
 
一比一原版(IIT毕业证)伊利诺伊理工大学毕业证成绩单
一比一原版(IIT毕业证)伊利诺伊理工大学毕业证成绩单一比一原版(IIT毕业证)伊利诺伊理工大学毕业证成绩单
一比一原版(IIT毕业证)伊利诺伊理工大学毕业证成绩单
ewymefz
 
一比一原版(UVic毕业证)维多利亚大学毕业证成绩单
一比一原版(UVic毕业证)维多利亚大学毕业证成绩单一比一原版(UVic毕业证)维多利亚大学毕业证成绩单
一比一原版(UVic毕业证)维多利亚大学毕业证成绩单
ukgaet
 
一比一原版(UMich毕业证)密歇根大学|安娜堡分校毕业证成绩单
一比一原版(UMich毕业证)密歇根大学|安娜堡分校毕业证成绩单一比一原版(UMich毕业证)密歇根大学|安娜堡分校毕业证成绩单
一比一原版(UMich毕业证)密歇根大学|安娜堡分校毕业证成绩单
ewymefz
 
The affect of service quality and online reviews on customer loyalty in the E...
The affect of service quality and online reviews on customer loyalty in the E...The affect of service quality and online reviews on customer loyalty in the E...
The affect of service quality and online reviews on customer loyalty in the E...
jerlynmaetalle
 
Criminal IP - Threat Hunting Webinar.pdf
Criminal IP - Threat Hunting Webinar.pdfCriminal IP - Threat Hunting Webinar.pdf
Criminal IP - Threat Hunting Webinar.pdf
Criminal IP
 
Chatty Kathy - UNC Bootcamp Final Project Presentation - Final Version - 5.23...
Chatty Kathy - UNC Bootcamp Final Project Presentation - Final Version - 5.23...Chatty Kathy - UNC Bootcamp Final Project Presentation - Final Version - 5.23...
Chatty Kathy - UNC Bootcamp Final Project Presentation - Final Version - 5.23...
John Andrews
 
一比一原版(CBU毕业证)不列颠海角大学毕业证成绩单
一比一原版(CBU毕业证)不列颠海角大学毕业证成绩单一比一原版(CBU毕业证)不列颠海角大学毕业证成绩单
一比一原版(CBU毕业证)不列颠海角大学毕业证成绩单
nscud
 
Algorithmic optimizations for Dynamic Levelwise PageRank (from STICD) : SHORT...
Algorithmic optimizations for Dynamic Levelwise PageRank (from STICD) : SHORT...Algorithmic optimizations for Dynamic Levelwise PageRank (from STICD) : SHORT...
Algorithmic optimizations for Dynamic Levelwise PageRank (from STICD) : SHORT...
Subhajit Sahu
 
Malana- Gimlet Market Analysis (Portfolio 2)
Malana- Gimlet Market Analysis (Portfolio 2)Malana- Gimlet Market Analysis (Portfolio 2)
Malana- Gimlet Market Analysis (Portfolio 2)
TravisMalana
 
一比一原版(UofM毕业证)明尼苏达大学毕业证成绩单
一比一原版(UofM毕业证)明尼苏达大学毕业证成绩单一比一原版(UofM毕业证)明尼苏达大学毕业证成绩单
一比一原版(UofM毕业证)明尼苏达大学毕业证成绩单
ewymefz
 
Sample_Global Non-invasive Prenatal Testing (NIPT) Market, 2019-2030.pdf
Sample_Global Non-invasive Prenatal Testing (NIPT) Market, 2019-2030.pdfSample_Global Non-invasive Prenatal Testing (NIPT) Market, 2019-2030.pdf
Sample_Global Non-invasive Prenatal Testing (NIPT) Market, 2019-2030.pdf
Linda486226
 
一比一原版(NYU毕业证)纽约大学毕业证成绩单
一比一原版(NYU毕业证)纽约大学毕业证成绩单一比一原版(NYU毕业证)纽约大学毕业证成绩单
一比一原版(NYU毕业证)纽约大学毕业证成绩单
ewymefz
 

Recently uploaded (20)

Tabula.io Cheatsheet: automate your data workflows
Tabula.io Cheatsheet: automate your data workflowsTabula.io Cheatsheet: automate your data workflows
Tabula.io Cheatsheet: automate your data workflows
 
Levelwise PageRank with Loop-Based Dead End Handling Strategy : SHORT REPORT ...
Levelwise PageRank with Loop-Based Dead End Handling Strategy : SHORT REPORT ...Levelwise PageRank with Loop-Based Dead End Handling Strategy : SHORT REPORT ...
Levelwise PageRank with Loop-Based Dead End Handling Strategy : SHORT REPORT ...
 
FP Growth Algorithm and its Applications
FP Growth Algorithm and its ApplicationsFP Growth Algorithm and its Applications
FP Growth Algorithm and its Applications
 
Jpolillo Amazon PPC - Bid Optimization Sample
Jpolillo Amazon PPC - Bid Optimization SampleJpolillo Amazon PPC - Bid Optimization Sample
Jpolillo Amazon PPC - Bid Optimization Sample
 
Adjusting primitives for graph : SHORT REPORT / NOTES
Adjusting primitives for graph : SHORT REPORT / NOTESAdjusting primitives for graph : SHORT REPORT / NOTES
Adjusting primitives for graph : SHORT REPORT / NOTES
 
Predicting Product Ad Campaign Performance: A Data Analysis Project Presentation
Predicting Product Ad Campaign Performance: A Data Analysis Project PresentationPredicting Product Ad Campaign Performance: A Data Analysis Project Presentation
Predicting Product Ad Campaign Performance: A Data Analysis Project Presentation
 
standardisation of garbhpala offhgfffghh
standardisation of garbhpala offhgfffghhstandardisation of garbhpala offhgfffghh
standardisation of garbhpala offhgfffghh
 
SOCRadar Germany 2024 Threat Landscape Report
SOCRadar Germany 2024 Threat Landscape ReportSOCRadar Germany 2024 Threat Landscape Report
SOCRadar Germany 2024 Threat Landscape Report
 
一比一原版(IIT毕业证)伊利诺伊理工大学毕业证成绩单
一比一原版(IIT毕业证)伊利诺伊理工大学毕业证成绩单一比一原版(IIT毕业证)伊利诺伊理工大学毕业证成绩单
一比一原版(IIT毕业证)伊利诺伊理工大学毕业证成绩单
 
一比一原版(UVic毕业证)维多利亚大学毕业证成绩单
一比一原版(UVic毕业证)维多利亚大学毕业证成绩单一比一原版(UVic毕业证)维多利亚大学毕业证成绩单
一比一原版(UVic毕业证)维多利亚大学毕业证成绩单
 
一比一原版(UMich毕业证)密歇根大学|安娜堡分校毕业证成绩单
一比一原版(UMich毕业证)密歇根大学|安娜堡分校毕业证成绩单一比一原版(UMich毕业证)密歇根大学|安娜堡分校毕业证成绩单
一比一原版(UMich毕业证)密歇根大学|安娜堡分校毕业证成绩单
 
The affect of service quality and online reviews on customer loyalty in the E...
The affect of service quality and online reviews on customer loyalty in the E...The affect of service quality and online reviews on customer loyalty in the E...
The affect of service quality and online reviews on customer loyalty in the E...
 
Criminal IP - Threat Hunting Webinar.pdf
Criminal IP - Threat Hunting Webinar.pdfCriminal IP - Threat Hunting Webinar.pdf
Criminal IP - Threat Hunting Webinar.pdf
 
Chatty Kathy - UNC Bootcamp Final Project Presentation - Final Version - 5.23...
Chatty Kathy - UNC Bootcamp Final Project Presentation - Final Version - 5.23...Chatty Kathy - UNC Bootcamp Final Project Presentation - Final Version - 5.23...
Chatty Kathy - UNC Bootcamp Final Project Presentation - Final Version - 5.23...
 
一比一原版(CBU毕业证)不列颠海角大学毕业证成绩单
一比一原版(CBU毕业证)不列颠海角大学毕业证成绩单一比一原版(CBU毕业证)不列颠海角大学毕业证成绩单
一比一原版(CBU毕业证)不列颠海角大学毕业证成绩单
 
Algorithmic optimizations for Dynamic Levelwise PageRank (from STICD) : SHORT...
Algorithmic optimizations for Dynamic Levelwise PageRank (from STICD) : SHORT...Algorithmic optimizations for Dynamic Levelwise PageRank (from STICD) : SHORT...
Algorithmic optimizations for Dynamic Levelwise PageRank (from STICD) : SHORT...
 
Malana- Gimlet Market Analysis (Portfolio 2)
Malana- Gimlet Market Analysis (Portfolio 2)Malana- Gimlet Market Analysis (Portfolio 2)
Malana- Gimlet Market Analysis (Portfolio 2)
 
一比一原版(UofM毕业证)明尼苏达大学毕业证成绩单
一比一原版(UofM毕业证)明尼苏达大学毕业证成绩单一比一原版(UofM毕业证)明尼苏达大学毕业证成绩单
一比一原版(UofM毕业证)明尼苏达大学毕业证成绩单
 
Sample_Global Non-invasive Prenatal Testing (NIPT) Market, 2019-2030.pdf
Sample_Global Non-invasive Prenatal Testing (NIPT) Market, 2019-2030.pdfSample_Global Non-invasive Prenatal Testing (NIPT) Market, 2019-2030.pdf
Sample_Global Non-invasive Prenatal Testing (NIPT) Market, 2019-2030.pdf
 
一比一原版(NYU毕业证)纽约大学毕业证成绩单
一比一原版(NYU毕业证)纽约大学毕业证成绩单一比一原版(NYU毕业证)纽约大学毕业证成绩单
一比一原版(NYU毕业证)纽约大学毕业证成绩单
 

Ai big dataconference_volodymyr getmanskyi colorization distance measuring

  • 1. Deep learning for satellite imagery colorization and distance measuring Lviv AI&BigData Day, November 4 Volodymyr Getmanskyi ELEKS data science team skype: paradoxx_xx
  • 2. Automatic image colorization is the task of adding colors to a new grayscale image without any user intervention. This problem is ill-posed in the sense that there is no unique colorization of a grayscale image without any prior knowledge. Indeed, many objects can have different colors: artificial objects, such as plastic objects which can have random colors, but also for natural objects such as tree leaves which can have various nuances of green and brown in different seasons, without significant change of shape.
  • 3. Most of the modern methods uses pretrained convolutional neural networks and extracts the information about features/objects from the image. Then there is a possibility to upscale them to original size and concatenating them all together. Also most of them uses some simplification to train the models to produce two color channels which can be concatenated with the grayscale input channel to produce the YUV/CIELUV image.
  • 4. The most obvious loss function is a Euclidean distance function between the network RGB output image and the true color RGB image (distance in UV space – from previous slide). We, however, notice that in numerous cases, the images colorized with such approach are mostly sepia- toned and muted in color. To understand why, consider a pixel that exists in a flower petal across multiple images that are identical, save for the color of the flower petals. Depending on the picture, this pixel can take on various tones of red, yellow, blue, and more. With a regression-based system that uses an loss function, the predicted pixel value that minimizes the loss for this particular pixel is the mean pixel value. Accordingly, the predicted pixel ends up being a mixture of the possible colors.
  • 5. Another problem is that pre-trained CNN weights, that was used for feature extraction, are from estimators that were trained on reсognizing different objects but not the objects from satellite images. Also all trained models for colorizing (with RGB/UV distance minimizing), were trained to restore colors for different objects but not for satellite images.
  • 6. We tried to test existing solutions but most of them are facing the problems that we mentioned earlier and demonstrates low performance: * http://demos.algorithmia.com/colorize-photos, http://www.colorizephoto.com/converter http://pinetools.com/colorize-image, https://github.com/richzhang/colorization
  • 7. After the first steps we found more efficient ways (with no or small sepia effects) and also try to train the estimator based on maps. Below you can see our basic results with adding color histograms and color mapping as additional map features:
  • 8. Below you can see our basic results with models training (black&white original, colorized, original): First iteration of our model (sepia warning): Weak trained model (1K iteration, 20K samples without augmentation):
  • 9. Measuring distance is a key tool in map reading and is especially useful for hikers and cyclists who want to measure how far they have travelled or how far they wish to go based on raw maps. BTW, such distance measuring is helpful for understanding map scale that is the first step for camera height evaluation (the distance from sea level to eye level) or altitude of the plane/drone above sea / terrain level (ASL). So the main task here is to get the scale. If you understand the map scale number that's great and you can still measure distances accurately. So based on such thoughts we decided to build the estimator that can measure distances on real unlabeled maps using some patterns like shadows, trees’ crowns, roofs and roads.
  • 10. For data gathering we used typical browser and google maps with ImageGrab and win32api python libraries the same as for colorization task (20K samples), but there are few image processing methods before the DL usage. So here we need to extract/recognize the scale that will be our labels. Scale is located on the right bottom of the image, so we need to detect the defining point from which we can move scale detection. Here we can simply slides the template image over the input image (as in cascade methods) and compares the template and patch of input image with some distance metric.
  • 11. After the successful detection we can extract two elements that we need for labeling: - Current measurement units (using simple OCR like tesseract and further rule-based engine) - Scale Line length (simply calculated the location of rows’ mode with continuous range of white columns)
  • 12. After the validation comparison with VGG architecture we choose AlexNet architecture with all convolutional layers frozen (transfer learning for efficient feature extraction): # Real-time data preprocessing img_prep = ImagePreprocessing() img_prep.add_featurewise_zero_center() img_prep.add_featurewise_stdnorm() # Real-time data augmentation img_aug = ImageAugmentation() img_aug.add_random_flip_leftright() img_aug.add_random_rotation(max_angle=135.)
  • 13. Results (examples from testset, mae≈25%): Image Side length Predicted length Image Side length Predicted length 168 m 213 m 45 m 69 m 1333 m 954 m 4970 m 4720 m
  • 14. Through our small experiment, we have checked the efficiency of using deep neural networks to colorize black and white satellite images the same as for measuring maps’ scale. So our conclusions are the following. In particular, formulating the maps colorization task as a classification problem can yield colorized maps that are arguably much more aesthetically-pleasing than those generated by a baseline regression-based model (with sepia), and thus shows much promise for further development. Also, redesigning the system around an adversarial network may improve results, since instead of focusing on minimizing the loss on a per-pixel deviation, the system would learn to generate/colorize pictures that compare well with real-world maps. From distance measuring approach we recommend to use similar feature extraction (transfer learning) that helps to feed the real objects (trees, roofs, roads) into distance result.
  • 15. Inspired by Technology. Driven by Value. Find us at eleks.com Have a question? Write to eleksinfo@eleks.com

Editor's Notes

  1. Use Ctrl+Shift+V (Cmd+Shift+V) to paste text (Thanks C.O.!) If your presentation is business confidential - keep “business confidential” mark. It other cases you can delete it.