SlideShare a Scribd company logo
1
Comparison of Fine-tuning and Extension
Strategies for Deep Convolutional Neural
Networks
Nikiforos Pittaras1, Foteini Markatopoulou1,2, Vasileios Mezaris1,
and Ioannis Patras2
1Information Technologies Institute / Centre for Research and Technology Hellas
2Queen Mary University of London
2
Problem
3
Typical solution
4
Typical solution
We evaluate
DCNN-based
approaches
5
Transfer learning
6
Literature review: Fine-tuning strategies
•Replacing the classification
layer with a new output
layer [2,5,18]
• The new layer is learned
from scratch
• All the other layers are fine-
tuned
7
•Re-initializing the last N
layers [1,9,18]
• The last N layers are
learned from scratch
• The first M-N layers are
fine-tuned with a low
learning rate [18] (or could
remain frozen [1,9])
Literature review: Fine-tuning strategies
8
•Extending the network by
one or more fully connected
layers [1,8,9,15]
Literature review: Fine-tuning strategies
9
FT3-ex: extension strategy
10
FT3-ex: extension strategy
Add one or more FC layers
before the classification layer
11
FT3-ex: extension strategy
Insert one or more FC layers
for each auxiliary classifier
12
FT3-ex: extension strategy
We use the output of the last
three layers as features to
train LR classifiers
13
FT3-ex: extension strategy
We also evaluate the direct
output of each network
14
Evaluation setup
Dataset: TRECVID SIN 2013
• 800 and 200 hours of internet archive videos for training and testing
• One keyframe per video shot
• Evaluated concepts: 38, Evaluation measure: MXinfAP
Dataset: PASCAL VOC 2012
• 5717 training, 5823 validation and 10991 test images
• Evaluation on the validation set instead of the original test set
• Evaluated concepts: 20, Evaluation measure: MAP
We fine-tuned 3 pre-trained ImageNet DCNNs:
• CaffeNet-1k, trained on 1000 ImageNet categories
• GoogLeNet-1k, trained on the same 1000 ImageNet categories
• GoogLeNet-5k, trained using 5055 ImageNet categories
15
Evaluation setup
For each pair of utilized network and fine-tuning strategy we evaluate:
• The direct output of the network
• Logistic regression (LR) classifiers trained on DCNN-based features
• One LR classifier per concept trained using the output of one layer
• The late-fused output (arithmetic mean) of LR classifiers trained using the
last three layers
We also evaluate the two auxiliary classifiers of the GoogLeNet-based networks
16
Preliminary experiments for
parameter selection
• Classification accuracy for the FT1-def strategy and CaffeNet-1k-60-SIN
• k: the learning rate multiplier of the pre-trained layers
• e: the number of training epochs
• The best accuracy per e is underlined; the globally best accuracy is bold and
underlined
• Improved accuracy for:
• Smaller learning rate values for the pre-trained layers
• Higher values for the training epochs
17
Experimental results – # target concepts
• Goal: Assess impact of number of target concepts
• Experiment: We fine-tune the network for either 60 or all 345 concepts; we
evaluate the same 38⊆60 concepts
• Conclusion: Fine-tuning a network for more concepts improves concept
detection accuracy
19
20
21
22
23
24
25
26
FT1-def
FT2-re1
FT2-re2
FT3-ex1-64
FT3-ex1-128
FT3-ex1-256
FT3-ex1-512
FT3-ex1-1024
FT3-ex1-2048
FT3-ex1-4096
FT3-ex2-64
FT3-ex2-128
FT3-ex2-256
FT3-ex2-512
FT3-ex2-1024
FT3-ex2-2048
FT3-ex2-4096
MXInfAP(%)
Fine-tuning strategies
TRECVID SIN Dataset
CaffeNet-1k-60-SIN
CaffeNet-1k-345-SIN
18
Experimental results – direct output
• Goal: Assess DCNNs as standalone classifiers (direct output)
• Experiment: Fine-tuning on the TRECVID SIN dataset
• Conclusion: FT3-ex1-64 and FT3-ex1-128 constitute the top-two methods
irrespective of the employed DCNN
14
16
18
20
22
24
26
FT1-def
FT2-re1
FT2-re2
FT3-ex1-64
FT3-ex1-128
FT3-ex1-256
FT3-ex1-512
FT3-ex1-1024
FT3-ex1-2048
FT3-ex1-4096
FT3-ex2-64
FT3-ex2-128
FT3-ex2-256
FT3-ex2-512
FT3-ex2-1024
FT3-ex2-2048
FT3-ex2-4096
MXInfAP(%)
Fine-tuning strategies
TRECVID SIN Dataset
CaffeNet-1k-SIN
GoogLeNet-1k-SIN
GoogLeNet-5k-SIN
19
Experimental results – direct output
• Goal: Assess DCNNs as standalone classifiers (direct output)
• Experiment: Fine-tuning on the PASCAL VOC dataset
• Conclusion: FT3-ex1-512 and FT3-ex1-1024 the best performing strategies for the
CaffeNet network
58
63
68
73
78
83
FT1-def
FT2-re1
FT2-re2
FT3-ex1-64
FT3-ex1-128
FT3-ex1-256
FT3-ex1-512
FT3-ex1-1024
FT3-ex1-2048
FT3-ex1-4096
FT3-ex2-64
FT3-ex2-128
FT3-ex2-256
FT3-ex2-512
FT3-ex2-1024
FT3-ex2-2048
FT3-ex2-4096
MAP(%)
Fine-tuning strategies
PASCAL VOC Dataset
CaffeNet-1k-VOC
GoogLeNet-1k-VOC
GoogLeNet-5k-VOC
20
Experimental results – direct output
• Goal: Assess DCNNs as standalone classifiers (direct output)
• Experiment: Fine-tuning on the PASCAL VOC dataset
• Conclusion: FT3-ex1-2048 and FT3-ex1-4096 the top-two methods for the
GoogLeNet-based networks
58
63
68
73
78
83
FT1-def
FT2-re1
FT2-re2
FT3-ex1-64
FT3-ex1-128
FT3-ex1-256
FT3-ex1-512
FT3-ex1-1024
FT3-ex1-2048
FT3-ex1-4096
FT3-ex2-64
FT3-ex2-128
FT3-ex2-256
FT3-ex2-512
FT3-ex2-1024
FT3-ex2-2048
FT3-ex2-4096
MAP(%)
Fine-tuning strategies
PASCAL VOC Dataset
CaffeNet-1k-VOC
GoogLeNet-1k-VOC
GoogLeNet-5k-VOC
21
Experimental results – direct output
• Main conclusions: FT3-ex strategy with one extension layer is always the best
solution
• The optimal dimension of the extension layer depends on the dataset and the
network architecture
22
Experimental results – DCNN features
• Goal: Assess DCNNs as feature generators (DCNN-based features)
• Experiment: LR concept detectors trained on the output of the last 3 layers and
fused in terms of arithmetic-mean
• Conclusion: FT3- ex1-512 in the top-five methods; FT3-ex2-64 is always among
the five worst fine-tuning methods
17
19
21
23
25
27
29
31
33
FT1-def
FT2-re1
FT2-re2
FT3-ex1-64
FT3-ex1-128
FT3-ex1-256
FT3-ex1-512
FT3-ex1-1024
FT3-ex1-2048
FT3-ex1-4096
FT3-ex2-64
FT3-ex2-128
FT3-ex2-256
FT3-ex2-512
FT3-ex2-1024
FT3-ex2-2048
FT3-ex2-4096
MXInfAP(%)
Fine-tuning strategies
TRECVID SIN Dataset
CaffeNet-1k-SIN
GoogLeNet-1k-SIN
GoogLeNet-5k-SIN
23
Experimental results – DCNN features
• The same conclusions hold for the PASCAL VOC Dataset
50
55
60
65
70
75
80
85
90
FT1-def
FT2-re1
FT2-re2
FT3-ex1-64
FT3-ex1-128
FT3-ex1-256
FT3-ex1-512
FT3-ex1-1024
FT3-ex1-2048
FT3-ex1-4096
FT3-ex2-64
FT3-ex2-128
FT3-ex2-256
FT3-ex2-512
FT3-ex2-1024
FT3-ex2-2048
FT3-ex2-4096
MAP(%)
Fine-tuning strategies
PASCAL VOC Dataset
CaffeNet-1k-VOC
GoogLeNet-1k-VOC
GoogLeNet-5k-VOC
24
Experimental results – DCNN features
• Main conclusions: FT3-ex strategy almost always outperforms the other two fine-
tuning strategies
• FT3-ex1-512 is in the top-five methods
• Additional conclusions: drawn from results presented in the paper
• Features extracted from the top layers are more accurate than layers
positioned lower in the network; the optimal layer varies, depending on the
target domain dataset
• Better to combine features extracted from many layers
• The presented results correspond to the fused output of the last 3 layers
25
Conclusions
• Extension strategy almost always outperforms all the other strategies
• Increase the depth with one fully-connected layer
• Fine-tune the rest of the layers
• DCNN-based features significantly outperform the direct output
• In a few cases the direct output works comparably well
• Choose based on the application that the DCNN will be used; e.g., real time
applications’ time and memory limitations
• Better to combine features extracted from many layers
26
References
[1] Campos, V., Salvador, A., Giro-i Nieto, X., Jou, B.: Diving deep into sentiment: understanding fine-tuned
CNNs for visual sentiment prediction. In: 1st International Workshop on Affect and Sentiment in Multimedia
(ASM 2015), pp. 57–62. ACM, Brisbane (2015)
[2] Chatfield, K., Simonyan, K., Vedaldi, A., Zisserman, A.: Return of the devil in the details: delving deep into
convolutional nets. In: British Machine Vision Conference (2014)
[5.] Girshick, R., Donahue, J., Darrell, T., Malik, J.: Rich feature hierarchies for accurate object detection and
semantic segmentation. In: Computer Vision and Pattern Recognition (CVPR 2014) (2014)
[8] Markatopoulou, F., et al.: ITI-CERTH participation in TRECVID 2015. In: TRECVID 2015 Workshop. NIST,
Gaithersburg (2015)
[9] Oquab, M., Bottou, L., Laptev, I., Sivic, J.: Learning and transferring mid-level image representations using
convolutional neural networks. In: Computer Vision and Pattern Recognition (CVPR 2014) (2014)
[15] Snoek, C., Fontijne, D., van de Sande, K.E., Stokman, H., et al.: Qualcomm Research and University of
Amsterdam at TRECVID 2015: recognizing concepts, objects, and events in video. In: TRECVID 2015 Workshop.
NIST, Gaithersburg (2015)
[18] Yosinski, J., Clune, J., Bengio, Y., Lipson, H.: How transferable are features in deep neural networks? CoRR
abs/1411.1792 (2014)
27
Thank you for your attention!
Questions?
More information and contact:
Dr. Vasileios Mezaris
bmezaris@iti.gr
http://www.iti.gr/~bmezaris

More Related Content

What's hot

Introduction to Model-Based Machine Learning
Introduction to Model-Based Machine LearningIntroduction to Model-Based Machine Learning
Introduction to Model-Based Machine Learning
Daniel Emaasit
 
MediaEval 2016 - UNIFESP Predicting Media Interestingness Task
MediaEval 2016 - UNIFESP Predicting Media Interestingness TaskMediaEval 2016 - UNIFESP Predicting Media Interestingness Task
MediaEval 2016 - UNIFESP Predicting Media Interestingness Task
multimediaeval
 
Convolutional neural networks deepa
Convolutional neural networks deepaConvolutional neural networks deepa
Convolutional neural networks deepa
deepa4466
 
CNN vs SIFT-based Visual Localization - Laura Leal-Taixé - UPC Barcelona 2018
CNN vs SIFT-based Visual Localization - Laura Leal-Taixé - UPC Barcelona 2018CNN vs SIFT-based Visual Localization - Laura Leal-Taixé - UPC Barcelona 2018
CNN vs SIFT-based Visual Localization - Laura Leal-Taixé - UPC Barcelona 2018
Universitat Politècnica de Catalunya
 
TRECVID 2016 Ad-hoc Video Seach task, CERTH-ITI
TRECVID 2016 Ad-hoc Video Seach task, CERTH-ITITRECVID 2016 Ad-hoc Video Seach task, CERTH-ITI
TRECVID 2016 Ad-hoc Video Seach task, CERTH-ITI
InVID Project
 
Scene classification using Convolutional Neural Networks - Jayani Withanawasam
Scene classification using Convolutional Neural Networks - Jayani WithanawasamScene classification using Convolutional Neural Networks - Jayani Withanawasam
Scene classification using Convolutional Neural Networks - Jayani Withanawasam
WithTheBest
 
Towards Set Learning and Prediction - Laura Leal-Taixe - UPC Barcelona 2018
Towards Set Learning and Prediction - Laura Leal-Taixe - UPC Barcelona 2018Towards Set Learning and Prediction - Laura Leal-Taixe - UPC Barcelona 2018
Towards Set Learning and Prediction - Laura Leal-Taixe - UPC Barcelona 2018
Universitat Politècnica de Catalunya
 
End-to-End Object Detection with Transformers
End-to-End Object Detection with TransformersEnd-to-End Object Detection with Transformers
End-to-End Object Detection with Transformers
Seunghyun Hwang
 
MediaEval 2016 - COSMIR and the OpenMIC Challenge: A Plan for Sustainable Mus...
MediaEval 2016 - COSMIR and the OpenMIC Challenge: A Plan for Sustainable Mus...MediaEval 2016 - COSMIR and the OpenMIC Challenge: A Plan for Sustainable Mus...
MediaEval 2016 - COSMIR and the OpenMIC Challenge: A Plan for Sustainable Mus...
multimediaeval
 
Life-long / Incremental Learning (DLAI D6L1 2017 UPC Deep Learning for Artifi...
Life-long / Incremental Learning (DLAI D6L1 2017 UPC Deep Learning for Artifi...Life-long / Incremental Learning (DLAI D6L1 2017 UPC Deep Learning for Artifi...
Life-long / Incremental Learning (DLAI D6L1 2017 UPC Deep Learning for Artifi...
Universitat Politècnica de Catalunya
 
Neuroevolution and deep learing
Neuroevolution and deep learing Neuroevolution and deep learing
Neuroevolution and deep learing
Accenture
 
Multiple Object Tracking - Laura Leal-Taixe - UPC Barcelona 2018
Multiple Object Tracking - Laura Leal-Taixe - UPC Barcelona 2018Multiple Object Tracking - Laura Leal-Taixe - UPC Barcelona 2018
Multiple Object Tracking - Laura Leal-Taixe - UPC Barcelona 2018
Universitat Politècnica de Catalunya
 
#4 Convolutional Neural Networks for Natural Language Processing
#4 Convolutional Neural Networks for Natural Language Processing#4 Convolutional Neural Networks for Natural Language Processing
#4 Convolutional Neural Networks for Natural Language Processing
Berlin Language Technology
 
MediaEval 2016 - HUCVL Predicting Interesting Key Frames with Deep Models
MediaEval 2016 - HUCVL Predicting Interesting Key Frames with Deep ModelsMediaEval 2016 - HUCVL Predicting Interesting Key Frames with Deep Models
MediaEval 2016 - HUCVL Predicting Interesting Key Frames with Deep Models
multimediaeval
 
Review: Incremental Few-shot Instance Segmentation [CDM]
Review: Incremental Few-shot Instance Segmentation [CDM]Review: Incremental Few-shot Instance Segmentation [CDM]
Review: Incremental Few-shot Instance Segmentation [CDM]
Dongmin Choi
 
How much position information do convolutional neural networks encode? review...
How much position information do convolutional neural networks encode? review...How much position information do convolutional neural networks encode? review...
How much position information do convolutional neural networks encode? review...
Dongmin Choi
 
MediaEval 2016 - ININ Submission to Zero Cost ASR Task
MediaEval 2016 - ININ Submission to Zero Cost ASR TaskMediaEval 2016 - ININ Submission to Zero Cost ASR Task
MediaEval 2016 - ININ Submission to Zero Cost ASR Task
multimediaeval
 
Deformable DETR Review [CDM]
Deformable DETR Review [CDM]Deformable DETR Review [CDM]
Deformable DETR Review [CDM]
Dongmin Choi
 
MediaEval 2016 - IR Evaluation: Putting the User Back in the Loop
MediaEval 2016 - IR Evaluation: Putting the User Back in the LoopMediaEval 2016 - IR Evaluation: Putting the User Back in the Loop
MediaEval 2016 - IR Evaluation: Putting the User Back in the Loop
multimediaeval
 
Transformer in Vision
Transformer in VisionTransformer in Vision
Transformer in Vision
Sangmin Woo
 

What's hot (20)

Introduction to Model-Based Machine Learning
Introduction to Model-Based Machine LearningIntroduction to Model-Based Machine Learning
Introduction to Model-Based Machine Learning
 
MediaEval 2016 - UNIFESP Predicting Media Interestingness Task
MediaEval 2016 - UNIFESP Predicting Media Interestingness TaskMediaEval 2016 - UNIFESP Predicting Media Interestingness Task
MediaEval 2016 - UNIFESP Predicting Media Interestingness Task
 
Convolutional neural networks deepa
Convolutional neural networks deepaConvolutional neural networks deepa
Convolutional neural networks deepa
 
CNN vs SIFT-based Visual Localization - Laura Leal-Taixé - UPC Barcelona 2018
CNN vs SIFT-based Visual Localization - Laura Leal-Taixé - UPC Barcelona 2018CNN vs SIFT-based Visual Localization - Laura Leal-Taixé - UPC Barcelona 2018
CNN vs SIFT-based Visual Localization - Laura Leal-Taixé - UPC Barcelona 2018
 
TRECVID 2016 Ad-hoc Video Seach task, CERTH-ITI
TRECVID 2016 Ad-hoc Video Seach task, CERTH-ITITRECVID 2016 Ad-hoc Video Seach task, CERTH-ITI
TRECVID 2016 Ad-hoc Video Seach task, CERTH-ITI
 
Scene classification using Convolutional Neural Networks - Jayani Withanawasam
Scene classification using Convolutional Neural Networks - Jayani WithanawasamScene classification using Convolutional Neural Networks - Jayani Withanawasam
Scene classification using Convolutional Neural Networks - Jayani Withanawasam
 
Towards Set Learning and Prediction - Laura Leal-Taixe - UPC Barcelona 2018
Towards Set Learning and Prediction - Laura Leal-Taixe - UPC Barcelona 2018Towards Set Learning and Prediction - Laura Leal-Taixe - UPC Barcelona 2018
Towards Set Learning and Prediction - Laura Leal-Taixe - UPC Barcelona 2018
 
End-to-End Object Detection with Transformers
End-to-End Object Detection with TransformersEnd-to-End Object Detection with Transformers
End-to-End Object Detection with Transformers
 
MediaEval 2016 - COSMIR and the OpenMIC Challenge: A Plan for Sustainable Mus...
MediaEval 2016 - COSMIR and the OpenMIC Challenge: A Plan for Sustainable Mus...MediaEval 2016 - COSMIR and the OpenMIC Challenge: A Plan for Sustainable Mus...
MediaEval 2016 - COSMIR and the OpenMIC Challenge: A Plan for Sustainable Mus...
 
Life-long / Incremental Learning (DLAI D6L1 2017 UPC Deep Learning for Artifi...
Life-long / Incremental Learning (DLAI D6L1 2017 UPC Deep Learning for Artifi...Life-long / Incremental Learning (DLAI D6L1 2017 UPC Deep Learning for Artifi...
Life-long / Incremental Learning (DLAI D6L1 2017 UPC Deep Learning for Artifi...
 
Neuroevolution and deep learing
Neuroevolution and deep learing Neuroevolution and deep learing
Neuroevolution and deep learing
 
Multiple Object Tracking - Laura Leal-Taixe - UPC Barcelona 2018
Multiple Object Tracking - Laura Leal-Taixe - UPC Barcelona 2018Multiple Object Tracking - Laura Leal-Taixe - UPC Barcelona 2018
Multiple Object Tracking - Laura Leal-Taixe - UPC Barcelona 2018
 
#4 Convolutional Neural Networks for Natural Language Processing
#4 Convolutional Neural Networks for Natural Language Processing#4 Convolutional Neural Networks for Natural Language Processing
#4 Convolutional Neural Networks for Natural Language Processing
 
MediaEval 2016 - HUCVL Predicting Interesting Key Frames with Deep Models
MediaEval 2016 - HUCVL Predicting Interesting Key Frames with Deep ModelsMediaEval 2016 - HUCVL Predicting Interesting Key Frames with Deep Models
MediaEval 2016 - HUCVL Predicting Interesting Key Frames with Deep Models
 
Review: Incremental Few-shot Instance Segmentation [CDM]
Review: Incremental Few-shot Instance Segmentation [CDM]Review: Incremental Few-shot Instance Segmentation [CDM]
Review: Incremental Few-shot Instance Segmentation [CDM]
 
How much position information do convolutional neural networks encode? review...
How much position information do convolutional neural networks encode? review...How much position information do convolutional neural networks encode? review...
How much position information do convolutional neural networks encode? review...
 
MediaEval 2016 - ININ Submission to Zero Cost ASR Task
MediaEval 2016 - ININ Submission to Zero Cost ASR TaskMediaEval 2016 - ININ Submission to Zero Cost ASR Task
MediaEval 2016 - ININ Submission to Zero Cost ASR Task
 
Deformable DETR Review [CDM]
Deformable DETR Review [CDM]Deformable DETR Review [CDM]
Deformable DETR Review [CDM]
 
MediaEval 2016 - IR Evaluation: Putting the User Back in the Loop
MediaEval 2016 - IR Evaluation: Putting the User Back in the LoopMediaEval 2016 - IR Evaluation: Putting the User Back in the Loop
MediaEval 2016 - IR Evaluation: Putting the User Back in the Loop
 
Transformer in Vision
Transformer in VisionTransformer in Vision
Transformer in Vision
 

Viewers also liked

InVID at the IPTC Autumn Meeting 2016
InVID at the IPTC Autumn Meeting 2016InVID at the IPTC Autumn Meeting 2016
InVID at the IPTC Autumn Meeting 2016
InVID Project
 
InVID verification application presentation at SMVW16
InVID verification application presentation at SMVW16InVID verification application presentation at SMVW16
InVID verification application presentation at SMVW16
InVID Project
 
InVID project presentation at SMVW16
InVID project presentation at SMVW16InVID project presentation at SMVW16
InVID project presentation at SMVW16
InVID Project
 
InVID platform presentation at SMVW16
InVID platform presentation at SMVW16InVID platform presentation at SMVW16
InVID platform presentation at SMVW16
InVID Project
 
InVID at Computational journalism workshop in Rennes, France
InVID at Computational journalism workshop in Rennes, FranceInVID at Computational journalism workshop in Rennes, France
InVID at Computational journalism workshop in Rennes, France
InVID Project
 
InVID Project Presentation 3rd release March 2016
InVID Project Presentation 3rd release March 2016InVID Project Presentation 3rd release March 2016
InVID Project Presentation 3rd release March 2016
InVID Project
 

Viewers also liked (6)

InVID at the IPTC Autumn Meeting 2016
InVID at the IPTC Autumn Meeting 2016InVID at the IPTC Autumn Meeting 2016
InVID at the IPTC Autumn Meeting 2016
 
InVID verification application presentation at SMVW16
InVID verification application presentation at SMVW16InVID verification application presentation at SMVW16
InVID verification application presentation at SMVW16
 
InVID project presentation at SMVW16
InVID project presentation at SMVW16InVID project presentation at SMVW16
InVID project presentation at SMVW16
 
InVID platform presentation at SMVW16
InVID platform presentation at SMVW16InVID platform presentation at SMVW16
InVID platform presentation at SMVW16
 
InVID at Computational journalism workshop in Rennes, France
InVID at Computational journalism workshop in Rennes, FranceInVID at Computational journalism workshop in Rennes, France
InVID at Computational journalism workshop in Rennes, France
 
InVID Project Presentation 3rd release March 2016
InVID Project Presentation 3rd release March 2016InVID Project Presentation 3rd release March 2016
InVID Project Presentation 3rd release March 2016
 

Similar to Comparison of Fine-tuning and Extension Strategies for Deep Convolutional Neural Networks

Deep Learning Initiative @ NECSTLab
Deep Learning Initiative @ NECSTLabDeep Learning Initiative @ NECSTLab
Deep Learning Initiative @ NECSTLab
NECST Lab @ Politecnico di Milano
 
"Quantizing Deep Networks for Efficient Inference at the Edge," a Presentatio...
"Quantizing Deep Networks for Efficient Inference at the Edge," a Presentatio..."Quantizing Deep Networks for Efficient Inference at the Edge," a Presentatio...
"Quantizing Deep Networks for Efficient Inference at the Edge," a Presentatio...
Edge AI and Vision Alliance
 
PR-187 : MorphNet: Fast & Simple Resource-Constrained Structure Learning of D...
PR-187 : MorphNet: Fast & Simple Resource-Constrained Structure Learning of D...PR-187 : MorphNet: Fast & Simple Resource-Constrained Structure Learning of D...
PR-187 : MorphNet: Fast & Simple Resource-Constrained Structure Learning of D...
Sunghoon Joo
 
CTF: Anomaly Detection in High-Dimensional Time Series with Coarse-to-Fine Mo...
CTF: Anomaly Detection in High-Dimensional Time Series with Coarse-to-Fine Mo...CTF: Anomaly Detection in High-Dimensional Time Series with Coarse-to-Fine Mo...
CTF: Anomaly Detection in High-Dimensional Time Series with Coarse-to-Fine Mo...
ssuser9357dd
 
Introduction to CNN Models: DenseNet & MobileNet
Introduction to CNN Models: DenseNet & MobileNetIntroduction to CNN Models: DenseNet & MobileNet
Introduction to CNN Models: DenseNet & MobileNet
KrishnakoumarC
 
powerpoint feb
powerpoint febpowerpoint feb
powerpoint feb
imu409
 
대용량 데이터 분석을 위한 병렬 Clustering 알고리즘 최적화
대용량 데이터 분석을 위한 병렬 Clustering 알고리즘 최적화대용량 데이터 분석을 위한 병렬 Clustering 알고리즘 최적화
대용량 데이터 분석을 위한 병렬 Clustering 알고리즘 최적화
NAVER Engineering
 
Program on Mathematical and Statistical Methods for Climate and the Earth Sys...
Program on Mathematical and Statistical Methods for Climate and the Earth Sys...Program on Mathematical and Statistical Methods for Climate and the Earth Sys...
Program on Mathematical and Statistical Methods for Climate and the Earth Sys...
The Statistical and Applied Mathematical Sciences Institute
 
[2020 CVPR Efficient DET paper review]
[2020 CVPR Efficient DET paper review][2020 CVPR Efficient DET paper review]
[2020 CVPR Efficient DET paper review]
taeseon ryu
 
Fractional step discriminant pruning
Fractional step discriminant pruningFractional step discriminant pruning
Fractional step discriminant pruning
VasileiosMezaris
 
Trackster Pruning at the CMS High-Granularity Calorimeter
Trackster Pruning at the CMS High-Granularity CalorimeterTrackster Pruning at the CMS High-Granularity Calorimeter
Trackster Pruning at the CMS High-Granularity Calorimeter
Yousef Fadila
 
Cvpr 2018 papers review (efficient computing)
Cvpr 2018 papers review (efficient computing)Cvpr 2018 papers review (efficient computing)
Cvpr 2018 papers review (efficient computing)
DonghyunKang12
 
Hanjun Dai, PhD Student, School of Computational Science and Engineering, Geo...
Hanjun Dai, PhD Student, School of Computational Science and Engineering, Geo...Hanjun Dai, PhD Student, School of Computational Science and Engineering, Geo...
Hanjun Dai, PhD Student, School of Computational Science and Engineering, Geo...
MLconf
 
Once-for-All: Train One Network and Specialize it for Efficient Deployment
 Once-for-All: Train One Network and Specialize it for Efficient Deployment Once-for-All: Train One Network and Specialize it for Efficient Deployment
Once-for-All: Train One Network and Specialize it for Efficient Deployment
taeseon ryu
 
AI optimizing HPC simulations (presentation from 6th EULAG Workshop)
AI optimizing HPC simulations (presentation from  6th EULAG Workshop)AI optimizing HPC simulations (presentation from  6th EULAG Workshop)
AI optimizing HPC simulations (presentation from 6th EULAG Workshop)
byteLAKE
 
CNN Dataflow implementation on FPGAs
CNN Dataflow implementation on FPGAsCNN Dataflow implementation on FPGAs
CNN Dataflow implementation on FPGAs
NECST Lab @ Politecnico di Milano
 
Unsupervised Learning: Clustering
Unsupervised Learning: Clustering Unsupervised Learning: Clustering
Unsupervised Learning: Clustering
Experfy
 
Detection focal loss 딥러닝 논문읽기 모임 발표자료
Detection focal loss 딥러닝 논문읽기 모임 발표자료Detection focal loss 딥러닝 논문읽기 모임 발표자료
Detection focal loss 딥러닝 논문읽기 모임 발표자료
taeseon ryu
 
FAST ALGORITHMS FOR UNSUPERVISED LEARNING IN LARGE DATA SETS
FAST ALGORITHMS FOR UNSUPERVISED LEARNING IN LARGE DATA SETSFAST ALGORITHMS FOR UNSUPERVISED LEARNING IN LARGE DATA SETS
FAST ALGORITHMS FOR UNSUPERVISED LEARNING IN LARGE DATA SETS
csandit
 
Cnn
CnnCnn

Similar to Comparison of Fine-tuning and Extension Strategies for Deep Convolutional Neural Networks (20)

Deep Learning Initiative @ NECSTLab
Deep Learning Initiative @ NECSTLabDeep Learning Initiative @ NECSTLab
Deep Learning Initiative @ NECSTLab
 
"Quantizing Deep Networks for Efficient Inference at the Edge," a Presentatio...
"Quantizing Deep Networks for Efficient Inference at the Edge," a Presentatio..."Quantizing Deep Networks for Efficient Inference at the Edge," a Presentatio...
"Quantizing Deep Networks for Efficient Inference at the Edge," a Presentatio...
 
PR-187 : MorphNet: Fast & Simple Resource-Constrained Structure Learning of D...
PR-187 : MorphNet: Fast & Simple Resource-Constrained Structure Learning of D...PR-187 : MorphNet: Fast & Simple Resource-Constrained Structure Learning of D...
PR-187 : MorphNet: Fast & Simple Resource-Constrained Structure Learning of D...
 
CTF: Anomaly Detection in High-Dimensional Time Series with Coarse-to-Fine Mo...
CTF: Anomaly Detection in High-Dimensional Time Series with Coarse-to-Fine Mo...CTF: Anomaly Detection in High-Dimensional Time Series with Coarse-to-Fine Mo...
CTF: Anomaly Detection in High-Dimensional Time Series with Coarse-to-Fine Mo...
 
Introduction to CNN Models: DenseNet & MobileNet
Introduction to CNN Models: DenseNet & MobileNetIntroduction to CNN Models: DenseNet & MobileNet
Introduction to CNN Models: DenseNet & MobileNet
 
powerpoint feb
powerpoint febpowerpoint feb
powerpoint feb
 
대용량 데이터 분석을 위한 병렬 Clustering 알고리즘 최적화
대용량 데이터 분석을 위한 병렬 Clustering 알고리즘 최적화대용량 데이터 분석을 위한 병렬 Clustering 알고리즘 최적화
대용량 데이터 분석을 위한 병렬 Clustering 알고리즘 최적화
 
Program on Mathematical and Statistical Methods for Climate and the Earth Sys...
Program on Mathematical and Statistical Methods for Climate and the Earth Sys...Program on Mathematical and Statistical Methods for Climate and the Earth Sys...
Program on Mathematical and Statistical Methods for Climate and the Earth Sys...
 
[2020 CVPR Efficient DET paper review]
[2020 CVPR Efficient DET paper review][2020 CVPR Efficient DET paper review]
[2020 CVPR Efficient DET paper review]
 
Fractional step discriminant pruning
Fractional step discriminant pruningFractional step discriminant pruning
Fractional step discriminant pruning
 
Trackster Pruning at the CMS High-Granularity Calorimeter
Trackster Pruning at the CMS High-Granularity CalorimeterTrackster Pruning at the CMS High-Granularity Calorimeter
Trackster Pruning at the CMS High-Granularity Calorimeter
 
Cvpr 2018 papers review (efficient computing)
Cvpr 2018 papers review (efficient computing)Cvpr 2018 papers review (efficient computing)
Cvpr 2018 papers review (efficient computing)
 
Hanjun Dai, PhD Student, School of Computational Science and Engineering, Geo...
Hanjun Dai, PhD Student, School of Computational Science and Engineering, Geo...Hanjun Dai, PhD Student, School of Computational Science and Engineering, Geo...
Hanjun Dai, PhD Student, School of Computational Science and Engineering, Geo...
 
Once-for-All: Train One Network and Specialize it for Efficient Deployment
 Once-for-All: Train One Network and Specialize it for Efficient Deployment Once-for-All: Train One Network and Specialize it for Efficient Deployment
Once-for-All: Train One Network and Specialize it for Efficient Deployment
 
AI optimizing HPC simulations (presentation from 6th EULAG Workshop)
AI optimizing HPC simulations (presentation from  6th EULAG Workshop)AI optimizing HPC simulations (presentation from  6th EULAG Workshop)
AI optimizing HPC simulations (presentation from 6th EULAG Workshop)
 
CNN Dataflow implementation on FPGAs
CNN Dataflow implementation on FPGAsCNN Dataflow implementation on FPGAs
CNN Dataflow implementation on FPGAs
 
Unsupervised Learning: Clustering
Unsupervised Learning: Clustering Unsupervised Learning: Clustering
Unsupervised Learning: Clustering
 
Detection focal loss 딥러닝 논문읽기 모임 발표자료
Detection focal loss 딥러닝 논문읽기 모임 발표자료Detection focal loss 딥러닝 논문읽기 모임 발표자료
Detection focal loss 딥러닝 논문읽기 모임 발표자료
 
FAST ALGORITHMS FOR UNSUPERVISED LEARNING IN LARGE DATA SETS
FAST ALGORITHMS FOR UNSUPERVISED LEARNING IN LARGE DATA SETSFAST ALGORITHMS FOR UNSUPERVISED LEARNING IN LARGE DATA SETS
FAST ALGORITHMS FOR UNSUPERVISED LEARNING IN LARGE DATA SETS
 
Cnn
CnnCnn
Cnn
 

More from InVID Project

InVID at IFCN Global Fact V
InVID at IFCN Global Fact VInVID at IFCN Global Fact V
InVID at IFCN Global Fact V
InVID Project
 
Presentation of the InVID technologies for image forensics analysis
Presentation of the InVID technologies for image forensics analysisPresentation of the InVID technologies for image forensics analysis
Presentation of the InVID technologies for image forensics analysis
InVID Project
 
Presentation of the InVID tool for video fragmentation and keyframe-based rev...
Presentation of the InVID tool for video fragmentation and keyframe-based rev...Presentation of the InVID tool for video fragmentation and keyframe-based rev...
Presentation of the InVID tool for video fragmentation and keyframe-based rev...
InVID Project
 
Presentation of the InVID project and verification technologies
Presentation of the InVID project and verification technologiesPresentation of the InVID project and verification technologies
Presentation of the InVID project and verification technologies
InVID Project
 
Presentation of the InVID verification technologies at IPTC 2018
Presentation of the InVID verification technologies at IPTC 2018Presentation of the InVID verification technologies at IPTC 2018
Presentation of the InVID verification technologies at IPTC 2018
InVID Project
 
IUT-Toulouse-InVID-2018
IUT-Toulouse-InVID-2018IUT-Toulouse-InVID-2018
IUT-Toulouse-InVID-2018
InVID Project
 
A state of the art in journalism about fake image and video detection
A state of the art in journalism about fake image and video detectionA state of the art in journalism about fake image and video detection
A state of the art in journalism about fake image and video detection
InVID Project
 
Presentation of the InVID tool for social media verification
Presentation of the InVID tool for social media verificationPresentation of the InVID tool for social media verification
Presentation of the InVID tool for social media verification
InVID Project
 
Presentation of the InVID tool for video fragmentation and reverse keyframe s...
Presentation of the InVID tool for video fragmentation and reverse keyframe s...Presentation of the InVID tool for video fragmentation and reverse keyframe s...
Presentation of the InVID tool for video fragmentation and reverse keyframe s...
InVID Project
 
Presentation of the InVID Verification Plugin
Presentation of the InVID Verification PluginPresentation of the InVID Verification Plugin
Presentation of the InVID Verification Plugin
InVID Project
 
Presentation of the InVID tools for image forensics analysis
Presentation of the InVID tools for image forensics analysisPresentation of the InVID tools for image forensics analysis
Presentation of the InVID tools for image forensics analysis
InVID Project
 
Overview of the InVID project
Overview of the InVID projectOverview of the InVID project
Overview of the InVID project
InVID Project
 
Second issue of the InVID newsletter
Second issue of the InVID newsletterSecond issue of the InVID newsletter
Second issue of the InVID newsletter
InVID Project
 
The InVID Plug-in: Web Video Verification on the Browser
The InVID Plug-in: Web Video Verification on the BrowserThe InVID Plug-in: Web Video Verification on the Browser
The InVID Plug-in: Web Video Verification on the Browser
InVID Project
 
Video Retrieval for Multimedia Verification of Breaking News on Social Networks
Video Retrieval for Multimedia Verification  of Breaking News on Social NetworksVideo Retrieval for Multimedia Verification  of Breaking News on Social Networks
Video Retrieval for Multimedia Verification of Breaking News on Social Networks
InVID Project
 
The InVID Vefirication Plugin at #IFRAExpo #DCXExpo
The InVID Vefirication Plugin at #IFRAExpo #DCXExpoThe InVID Vefirication Plugin at #IFRAExpo #DCXExpo
The InVID Vefirication Plugin at #IFRAExpo #DCXExpo
InVID Project
 
The InVID Vefirication Plugin at DisinfoLab
The InVID Vefirication Plugin at DisinfoLabThe InVID Vefirication Plugin at DisinfoLab
The InVID Vefirication Plugin at DisinfoLab
InVID Project
 

More from InVID Project (17)

InVID at IFCN Global Fact V
InVID at IFCN Global Fact VInVID at IFCN Global Fact V
InVID at IFCN Global Fact V
 
Presentation of the InVID technologies for image forensics analysis
Presentation of the InVID technologies for image forensics analysisPresentation of the InVID technologies for image forensics analysis
Presentation of the InVID technologies for image forensics analysis
 
Presentation of the InVID tool for video fragmentation and keyframe-based rev...
Presentation of the InVID tool for video fragmentation and keyframe-based rev...Presentation of the InVID tool for video fragmentation and keyframe-based rev...
Presentation of the InVID tool for video fragmentation and keyframe-based rev...
 
Presentation of the InVID project and verification technologies
Presentation of the InVID project and verification technologiesPresentation of the InVID project and verification technologies
Presentation of the InVID project and verification technologies
 
Presentation of the InVID verification technologies at IPTC 2018
Presentation of the InVID verification technologies at IPTC 2018Presentation of the InVID verification technologies at IPTC 2018
Presentation of the InVID verification technologies at IPTC 2018
 
IUT-Toulouse-InVID-2018
IUT-Toulouse-InVID-2018IUT-Toulouse-InVID-2018
IUT-Toulouse-InVID-2018
 
A state of the art in journalism about fake image and video detection
A state of the art in journalism about fake image and video detectionA state of the art in journalism about fake image and video detection
A state of the art in journalism about fake image and video detection
 
Presentation of the InVID tool for social media verification
Presentation of the InVID tool for social media verificationPresentation of the InVID tool for social media verification
Presentation of the InVID tool for social media verification
 
Presentation of the InVID tool for video fragmentation and reverse keyframe s...
Presentation of the InVID tool for video fragmentation and reverse keyframe s...Presentation of the InVID tool for video fragmentation and reverse keyframe s...
Presentation of the InVID tool for video fragmentation and reverse keyframe s...
 
Presentation of the InVID Verification Plugin
Presentation of the InVID Verification PluginPresentation of the InVID Verification Plugin
Presentation of the InVID Verification Plugin
 
Presentation of the InVID tools for image forensics analysis
Presentation of the InVID tools for image forensics analysisPresentation of the InVID tools for image forensics analysis
Presentation of the InVID tools for image forensics analysis
 
Overview of the InVID project
Overview of the InVID projectOverview of the InVID project
Overview of the InVID project
 
Second issue of the InVID newsletter
Second issue of the InVID newsletterSecond issue of the InVID newsletter
Second issue of the InVID newsletter
 
The InVID Plug-in: Web Video Verification on the Browser
The InVID Plug-in: Web Video Verification on the BrowserThe InVID Plug-in: Web Video Verification on the Browser
The InVID Plug-in: Web Video Verification on the Browser
 
Video Retrieval for Multimedia Verification of Breaking News on Social Networks
Video Retrieval for Multimedia Verification  of Breaking News on Social NetworksVideo Retrieval for Multimedia Verification  of Breaking News on Social Networks
Video Retrieval for Multimedia Verification of Breaking News on Social Networks
 
The InVID Vefirication Plugin at #IFRAExpo #DCXExpo
The InVID Vefirication Plugin at #IFRAExpo #DCXExpoThe InVID Vefirication Plugin at #IFRAExpo #DCXExpo
The InVID Vefirication Plugin at #IFRAExpo #DCXExpo
 
The InVID Vefirication Plugin at DisinfoLab
The InVID Vefirication Plugin at DisinfoLabThe InVID Vefirication Plugin at DisinfoLab
The InVID Vefirication Plugin at DisinfoLab
 

Recently uploaded

How to Build a Profitable IoT Product.pptx
How to Build a Profitable IoT Product.pptxHow to Build a Profitable IoT Product.pptx
How to Build a Profitable IoT Product.pptx
Adam Dunkels
 
Girls call Kolkata 👀 XXXXXXXXXXX 👀 Rs.9.5 K Cash Payment With Room Delivery
Girls call Kolkata 👀 XXXXXXXXXXX 👀 Rs.9.5 K Cash Payment With Room Delivery Girls call Kolkata 👀 XXXXXXXXXXX 👀 Rs.9.5 K Cash Payment With Room Delivery
Girls call Kolkata 👀 XXXXXXXXXXX 👀 Rs.9.5 K Cash Payment With Room Delivery
sunilverma7884
 
WhatsApp Spy Online Trackers and Monitoring Apps
WhatsApp Spy Online Trackers and Monitoring AppsWhatsApp Spy Online Trackers and Monitoring Apps
WhatsApp Spy Online Trackers and Monitoring Apps
HackersList
 
Feature sql server terbaru performance.pptx
Feature sql server terbaru performance.pptxFeature sql server terbaru performance.pptx
Feature sql server terbaru performance.pptx
ssuser1915fe1
 
High Profile Girls call Service Pune 000XX00000 Provide Best And Top Girl Ser...
High Profile Girls call Service Pune 000XX00000 Provide Best And Top Girl Ser...High Profile Girls call Service Pune 000XX00000 Provide Best And Top Girl Ser...
High Profile Girls call Service Pune 000XX00000 Provide Best And Top Girl Ser...
bhumivarma35300
 
"Mastering Graphic Design: Essential Tips and Tricks for Beginners and Profes...
"Mastering Graphic Design: Essential Tips and Tricks for Beginners and Profes..."Mastering Graphic Design: Essential Tips and Tricks for Beginners and Profes...
"Mastering Graphic Design: Essential Tips and Tricks for Beginners and Profes...
Anant Gupta
 
Acumatica vs. Sage Intacct vs. NetSuite _ NOW CFO.pdf
Acumatica vs. Sage Intacct vs. NetSuite _ NOW CFO.pdfAcumatica vs. Sage Intacct vs. NetSuite _ NOW CFO.pdf
Acumatica vs. Sage Intacct vs. NetSuite _ NOW CFO.pdf
BrainSell Technologies
 
Recent Advancements in the NIST-JARVIS Infrastructure
Recent Advancements in the NIST-JARVIS InfrastructureRecent Advancements in the NIST-JARVIS Infrastructure
Recent Advancements in the NIST-JARVIS Infrastructure
KAMAL CHOUDHARY
 
Evolution of iPaaS - simplify IT workloads to provide a unified view of data...
Evolution of iPaaS - simplify IT workloads to provide a unified view of  data...Evolution of iPaaS - simplify IT workloads to provide a unified view of  data...
Evolution of iPaaS - simplify IT workloads to provide a unified view of data...
Torry Harris
 
(CISOPlatform Summit & SACON 2024) Digital Personal Data Protection Act.pdf
(CISOPlatform Summit & SACON 2024) Digital Personal Data Protection Act.pdf(CISOPlatform Summit & SACON 2024) Digital Personal Data Protection Act.pdf
(CISOPlatform Summit & SACON 2024) Digital Personal Data Protection Act.pdf
Priyanka Aash
 
How RPA Help in the Transportation and Logistics Industry.pptx
How RPA Help in the Transportation and Logistics Industry.pptxHow RPA Help in the Transportation and Logistics Industry.pptx
How RPA Help in the Transportation and Logistics Industry.pptx
SynapseIndia
 
Best Practices for Effectively Running dbt in Airflow.pdf
Best Practices for Effectively Running dbt in Airflow.pdfBest Practices for Effectively Running dbt in Airflow.pdf
Best Practices for Effectively Running dbt in Airflow.pdf
Tatiana Al-Chueyr
 
CiscoIconsLibrary cours de réseau VLAN.ppt
CiscoIconsLibrary cours de réseau VLAN.pptCiscoIconsLibrary cours de réseau VLAN.ppt
CiscoIconsLibrary cours de réseau VLAN.ppt
moinahousna
 
July Patch Tuesday
July Patch TuesdayJuly Patch Tuesday
July Patch Tuesday
Ivanti
 
WPRiders Company Presentation Slide Deck
WPRiders Company Presentation Slide DeckWPRiders Company Presentation Slide Deck
WPRiders Company Presentation Slide Deck
Lidia A.
 
leewayhertz.com-AI agents for healthcare Applications benefits and implementa...
leewayhertz.com-AI agents for healthcare Applications benefits and implementa...leewayhertz.com-AI agents for healthcare Applications benefits and implementa...
leewayhertz.com-AI agents for healthcare Applications benefits and implementa...
alexjohnson7307
 
WhatsApp Image 2024-03-27 at 08.19.52_bfd93109.pdf
WhatsApp Image 2024-03-27 at 08.19.52_bfd93109.pdfWhatsApp Image 2024-03-27 at 08.19.52_bfd93109.pdf
WhatsApp Image 2024-03-27 at 08.19.52_bfd93109.pdf
ArgaBisma
 
(CISOPlatform Summit & SACON 2024) Keynote _ Power Digital Identities With AI...
(CISOPlatform Summit & SACON 2024) Keynote _ Power Digital Identities With AI...(CISOPlatform Summit & SACON 2024) Keynote _ Power Digital Identities With AI...
(CISOPlatform Summit & SACON 2024) Keynote _ Power Digital Identities With AI...
Priyanka Aash
 
High Profile Girls Call ServiCe Hyderabad 0000000000 Tanisha Best High Class ...
High Profile Girls Call ServiCe Hyderabad 0000000000 Tanisha Best High Class ...High Profile Girls Call ServiCe Hyderabad 0000000000 Tanisha Best High Class ...
High Profile Girls Call ServiCe Hyderabad 0000000000 Tanisha Best High Class ...
aslasdfmkhan4750
 
RPA In Healthcare Benefits, Use Case, Trend And Challenges 2024.pptx
RPA In Healthcare Benefits, Use Case, Trend And Challenges 2024.pptxRPA In Healthcare Benefits, Use Case, Trend And Challenges 2024.pptx
RPA In Healthcare Benefits, Use Case, Trend And Challenges 2024.pptx
SynapseIndia
 

Recently uploaded (20)

How to Build a Profitable IoT Product.pptx
How to Build a Profitable IoT Product.pptxHow to Build a Profitable IoT Product.pptx
How to Build a Profitable IoT Product.pptx
 
Girls call Kolkata 👀 XXXXXXXXXXX 👀 Rs.9.5 K Cash Payment With Room Delivery
Girls call Kolkata 👀 XXXXXXXXXXX 👀 Rs.9.5 K Cash Payment With Room Delivery Girls call Kolkata 👀 XXXXXXXXXXX 👀 Rs.9.5 K Cash Payment With Room Delivery
Girls call Kolkata 👀 XXXXXXXXXXX 👀 Rs.9.5 K Cash Payment With Room Delivery
 
WhatsApp Spy Online Trackers and Monitoring Apps
WhatsApp Spy Online Trackers and Monitoring AppsWhatsApp Spy Online Trackers and Monitoring Apps
WhatsApp Spy Online Trackers and Monitoring Apps
 
Feature sql server terbaru performance.pptx
Feature sql server terbaru performance.pptxFeature sql server terbaru performance.pptx
Feature sql server terbaru performance.pptx
 
High Profile Girls call Service Pune 000XX00000 Provide Best And Top Girl Ser...
High Profile Girls call Service Pune 000XX00000 Provide Best And Top Girl Ser...High Profile Girls call Service Pune 000XX00000 Provide Best And Top Girl Ser...
High Profile Girls call Service Pune 000XX00000 Provide Best And Top Girl Ser...
 
"Mastering Graphic Design: Essential Tips and Tricks for Beginners and Profes...
"Mastering Graphic Design: Essential Tips and Tricks for Beginners and Profes..."Mastering Graphic Design: Essential Tips and Tricks for Beginners and Profes...
"Mastering Graphic Design: Essential Tips and Tricks for Beginners and Profes...
 
Acumatica vs. Sage Intacct vs. NetSuite _ NOW CFO.pdf
Acumatica vs. Sage Intacct vs. NetSuite _ NOW CFO.pdfAcumatica vs. Sage Intacct vs. NetSuite _ NOW CFO.pdf
Acumatica vs. Sage Intacct vs. NetSuite _ NOW CFO.pdf
 
Recent Advancements in the NIST-JARVIS Infrastructure
Recent Advancements in the NIST-JARVIS InfrastructureRecent Advancements in the NIST-JARVIS Infrastructure
Recent Advancements in the NIST-JARVIS Infrastructure
 
Evolution of iPaaS - simplify IT workloads to provide a unified view of data...
Evolution of iPaaS - simplify IT workloads to provide a unified view of  data...Evolution of iPaaS - simplify IT workloads to provide a unified view of  data...
Evolution of iPaaS - simplify IT workloads to provide a unified view of data...
 
(CISOPlatform Summit & SACON 2024) Digital Personal Data Protection Act.pdf
(CISOPlatform Summit & SACON 2024) Digital Personal Data Protection Act.pdf(CISOPlatform Summit & SACON 2024) Digital Personal Data Protection Act.pdf
(CISOPlatform Summit & SACON 2024) Digital Personal Data Protection Act.pdf
 
How RPA Help in the Transportation and Logistics Industry.pptx
How RPA Help in the Transportation and Logistics Industry.pptxHow RPA Help in the Transportation and Logistics Industry.pptx
How RPA Help in the Transportation and Logistics Industry.pptx
 
Best Practices for Effectively Running dbt in Airflow.pdf
Best Practices for Effectively Running dbt in Airflow.pdfBest Practices for Effectively Running dbt in Airflow.pdf
Best Practices for Effectively Running dbt in Airflow.pdf
 
CiscoIconsLibrary cours de réseau VLAN.ppt
CiscoIconsLibrary cours de réseau VLAN.pptCiscoIconsLibrary cours de réseau VLAN.ppt
CiscoIconsLibrary cours de réseau VLAN.ppt
 
July Patch Tuesday
July Patch TuesdayJuly Patch Tuesday
July Patch Tuesday
 
WPRiders Company Presentation Slide Deck
WPRiders Company Presentation Slide DeckWPRiders Company Presentation Slide Deck
WPRiders Company Presentation Slide Deck
 
leewayhertz.com-AI agents for healthcare Applications benefits and implementa...
leewayhertz.com-AI agents for healthcare Applications benefits and implementa...leewayhertz.com-AI agents for healthcare Applications benefits and implementa...
leewayhertz.com-AI agents for healthcare Applications benefits and implementa...
 
WhatsApp Image 2024-03-27 at 08.19.52_bfd93109.pdf
WhatsApp Image 2024-03-27 at 08.19.52_bfd93109.pdfWhatsApp Image 2024-03-27 at 08.19.52_bfd93109.pdf
WhatsApp Image 2024-03-27 at 08.19.52_bfd93109.pdf
 
(CISOPlatform Summit & SACON 2024) Keynote _ Power Digital Identities With AI...
(CISOPlatform Summit & SACON 2024) Keynote _ Power Digital Identities With AI...(CISOPlatform Summit & SACON 2024) Keynote _ Power Digital Identities With AI...
(CISOPlatform Summit & SACON 2024) Keynote _ Power Digital Identities With AI...
 
High Profile Girls Call ServiCe Hyderabad 0000000000 Tanisha Best High Class ...
High Profile Girls Call ServiCe Hyderabad 0000000000 Tanisha Best High Class ...High Profile Girls Call ServiCe Hyderabad 0000000000 Tanisha Best High Class ...
High Profile Girls Call ServiCe Hyderabad 0000000000 Tanisha Best High Class ...
 
RPA In Healthcare Benefits, Use Case, Trend And Challenges 2024.pptx
RPA In Healthcare Benefits, Use Case, Trend And Challenges 2024.pptxRPA In Healthcare Benefits, Use Case, Trend And Challenges 2024.pptx
RPA In Healthcare Benefits, Use Case, Trend And Challenges 2024.pptx
 

Comparison of Fine-tuning and Extension Strategies for Deep Convolutional Neural Networks

  • 1. 1 Comparison of Fine-tuning and Extension Strategies for Deep Convolutional Neural Networks Nikiforos Pittaras1, Foteini Markatopoulou1,2, Vasileios Mezaris1, and Ioannis Patras2 1Information Technologies Institute / Centre for Research and Technology Hellas 2Queen Mary University of London
  • 6. 6 Literature review: Fine-tuning strategies •Replacing the classification layer with a new output layer [2,5,18] • The new layer is learned from scratch • All the other layers are fine- tuned
  • 7. 7 •Re-initializing the last N layers [1,9,18] • The last N layers are learned from scratch • The first M-N layers are fine-tuned with a low learning rate [18] (or could remain frozen [1,9]) Literature review: Fine-tuning strategies
  • 8. 8 •Extending the network by one or more fully connected layers [1,8,9,15] Literature review: Fine-tuning strategies
  • 10. 10 FT3-ex: extension strategy Add one or more FC layers before the classification layer
  • 11. 11 FT3-ex: extension strategy Insert one or more FC layers for each auxiliary classifier
  • 12. 12 FT3-ex: extension strategy We use the output of the last three layers as features to train LR classifiers
  • 13. 13 FT3-ex: extension strategy We also evaluate the direct output of each network
  • 14. 14 Evaluation setup Dataset: TRECVID SIN 2013 • 800 and 200 hours of internet archive videos for training and testing • One keyframe per video shot • Evaluated concepts: 38, Evaluation measure: MXinfAP Dataset: PASCAL VOC 2012 • 5717 training, 5823 validation and 10991 test images • Evaluation on the validation set instead of the original test set • Evaluated concepts: 20, Evaluation measure: MAP We fine-tuned 3 pre-trained ImageNet DCNNs: • CaffeNet-1k, trained on 1000 ImageNet categories • GoogLeNet-1k, trained on the same 1000 ImageNet categories • GoogLeNet-5k, trained using 5055 ImageNet categories
  • 15. 15 Evaluation setup For each pair of utilized network and fine-tuning strategy we evaluate: • The direct output of the network • Logistic regression (LR) classifiers trained on DCNN-based features • One LR classifier per concept trained using the output of one layer • The late-fused output (arithmetic mean) of LR classifiers trained using the last three layers We also evaluate the two auxiliary classifiers of the GoogLeNet-based networks
  • 16. 16 Preliminary experiments for parameter selection • Classification accuracy for the FT1-def strategy and CaffeNet-1k-60-SIN • k: the learning rate multiplier of the pre-trained layers • e: the number of training epochs • The best accuracy per e is underlined; the globally best accuracy is bold and underlined • Improved accuracy for: • Smaller learning rate values for the pre-trained layers • Higher values for the training epochs
  • 17. 17 Experimental results – # target concepts • Goal: Assess impact of number of target concepts • Experiment: We fine-tune the network for either 60 or all 345 concepts; we evaluate the same 38⊆60 concepts • Conclusion: Fine-tuning a network for more concepts improves concept detection accuracy 19 20 21 22 23 24 25 26 FT1-def FT2-re1 FT2-re2 FT3-ex1-64 FT3-ex1-128 FT3-ex1-256 FT3-ex1-512 FT3-ex1-1024 FT3-ex1-2048 FT3-ex1-4096 FT3-ex2-64 FT3-ex2-128 FT3-ex2-256 FT3-ex2-512 FT3-ex2-1024 FT3-ex2-2048 FT3-ex2-4096 MXInfAP(%) Fine-tuning strategies TRECVID SIN Dataset CaffeNet-1k-60-SIN CaffeNet-1k-345-SIN
  • 18. 18 Experimental results – direct output • Goal: Assess DCNNs as standalone classifiers (direct output) • Experiment: Fine-tuning on the TRECVID SIN dataset • Conclusion: FT3-ex1-64 and FT3-ex1-128 constitute the top-two methods irrespective of the employed DCNN 14 16 18 20 22 24 26 FT1-def FT2-re1 FT2-re2 FT3-ex1-64 FT3-ex1-128 FT3-ex1-256 FT3-ex1-512 FT3-ex1-1024 FT3-ex1-2048 FT3-ex1-4096 FT3-ex2-64 FT3-ex2-128 FT3-ex2-256 FT3-ex2-512 FT3-ex2-1024 FT3-ex2-2048 FT3-ex2-4096 MXInfAP(%) Fine-tuning strategies TRECVID SIN Dataset CaffeNet-1k-SIN GoogLeNet-1k-SIN GoogLeNet-5k-SIN
  • 19. 19 Experimental results – direct output • Goal: Assess DCNNs as standalone classifiers (direct output) • Experiment: Fine-tuning on the PASCAL VOC dataset • Conclusion: FT3-ex1-512 and FT3-ex1-1024 the best performing strategies for the CaffeNet network 58 63 68 73 78 83 FT1-def FT2-re1 FT2-re2 FT3-ex1-64 FT3-ex1-128 FT3-ex1-256 FT3-ex1-512 FT3-ex1-1024 FT3-ex1-2048 FT3-ex1-4096 FT3-ex2-64 FT3-ex2-128 FT3-ex2-256 FT3-ex2-512 FT3-ex2-1024 FT3-ex2-2048 FT3-ex2-4096 MAP(%) Fine-tuning strategies PASCAL VOC Dataset CaffeNet-1k-VOC GoogLeNet-1k-VOC GoogLeNet-5k-VOC
  • 20. 20 Experimental results – direct output • Goal: Assess DCNNs as standalone classifiers (direct output) • Experiment: Fine-tuning on the PASCAL VOC dataset • Conclusion: FT3-ex1-2048 and FT3-ex1-4096 the top-two methods for the GoogLeNet-based networks 58 63 68 73 78 83 FT1-def FT2-re1 FT2-re2 FT3-ex1-64 FT3-ex1-128 FT3-ex1-256 FT3-ex1-512 FT3-ex1-1024 FT3-ex1-2048 FT3-ex1-4096 FT3-ex2-64 FT3-ex2-128 FT3-ex2-256 FT3-ex2-512 FT3-ex2-1024 FT3-ex2-2048 FT3-ex2-4096 MAP(%) Fine-tuning strategies PASCAL VOC Dataset CaffeNet-1k-VOC GoogLeNet-1k-VOC GoogLeNet-5k-VOC
  • 21. 21 Experimental results – direct output • Main conclusions: FT3-ex strategy with one extension layer is always the best solution • The optimal dimension of the extension layer depends on the dataset and the network architecture
  • 22. 22 Experimental results – DCNN features • Goal: Assess DCNNs as feature generators (DCNN-based features) • Experiment: LR concept detectors trained on the output of the last 3 layers and fused in terms of arithmetic-mean • Conclusion: FT3- ex1-512 in the top-five methods; FT3-ex2-64 is always among the five worst fine-tuning methods 17 19 21 23 25 27 29 31 33 FT1-def FT2-re1 FT2-re2 FT3-ex1-64 FT3-ex1-128 FT3-ex1-256 FT3-ex1-512 FT3-ex1-1024 FT3-ex1-2048 FT3-ex1-4096 FT3-ex2-64 FT3-ex2-128 FT3-ex2-256 FT3-ex2-512 FT3-ex2-1024 FT3-ex2-2048 FT3-ex2-4096 MXInfAP(%) Fine-tuning strategies TRECVID SIN Dataset CaffeNet-1k-SIN GoogLeNet-1k-SIN GoogLeNet-5k-SIN
  • 23. 23 Experimental results – DCNN features • The same conclusions hold for the PASCAL VOC Dataset 50 55 60 65 70 75 80 85 90 FT1-def FT2-re1 FT2-re2 FT3-ex1-64 FT3-ex1-128 FT3-ex1-256 FT3-ex1-512 FT3-ex1-1024 FT3-ex1-2048 FT3-ex1-4096 FT3-ex2-64 FT3-ex2-128 FT3-ex2-256 FT3-ex2-512 FT3-ex2-1024 FT3-ex2-2048 FT3-ex2-4096 MAP(%) Fine-tuning strategies PASCAL VOC Dataset CaffeNet-1k-VOC GoogLeNet-1k-VOC GoogLeNet-5k-VOC
  • 24. 24 Experimental results – DCNN features • Main conclusions: FT3-ex strategy almost always outperforms the other two fine- tuning strategies • FT3-ex1-512 is in the top-five methods • Additional conclusions: drawn from results presented in the paper • Features extracted from the top layers are more accurate than layers positioned lower in the network; the optimal layer varies, depending on the target domain dataset • Better to combine features extracted from many layers • The presented results correspond to the fused output of the last 3 layers
  • 25. 25 Conclusions • Extension strategy almost always outperforms all the other strategies • Increase the depth with one fully-connected layer • Fine-tune the rest of the layers • DCNN-based features significantly outperform the direct output • In a few cases the direct output works comparably well • Choose based on the application that the DCNN will be used; e.g., real time applications’ time and memory limitations • Better to combine features extracted from many layers
  • 26. 26 References [1] Campos, V., Salvador, A., Giro-i Nieto, X., Jou, B.: Diving deep into sentiment: understanding fine-tuned CNNs for visual sentiment prediction. In: 1st International Workshop on Affect and Sentiment in Multimedia (ASM 2015), pp. 57–62. ACM, Brisbane (2015) [2] Chatfield, K., Simonyan, K., Vedaldi, A., Zisserman, A.: Return of the devil in the details: delving deep into convolutional nets. In: British Machine Vision Conference (2014) [5.] Girshick, R., Donahue, J., Darrell, T., Malik, J.: Rich feature hierarchies for accurate object detection and semantic segmentation. In: Computer Vision and Pattern Recognition (CVPR 2014) (2014) [8] Markatopoulou, F., et al.: ITI-CERTH participation in TRECVID 2015. In: TRECVID 2015 Workshop. NIST, Gaithersburg (2015) [9] Oquab, M., Bottou, L., Laptev, I., Sivic, J.: Learning and transferring mid-level image representations using convolutional neural networks. In: Computer Vision and Pattern Recognition (CVPR 2014) (2014) [15] Snoek, C., Fontijne, D., van de Sande, K.E., Stokman, H., et al.: Qualcomm Research and University of Amsterdam at TRECVID 2015: recognizing concepts, objects, and events in video. In: TRECVID 2015 Workshop. NIST, Gaithersburg (2015) [18] Yosinski, J., Clune, J., Bengio, Y., Lipson, H.: How transferable are features in deep neural networks? CoRR abs/1411.1792 (2014)
  • 27. 27 Thank you for your attention! Questions? More information and contact: Dr. Vasileios Mezaris bmezaris@iti.gr http://www.iti.gr/~bmezaris