2. 2
Carmine Paolino
MSc in Artificial Intelligence at Vrije Universiteit Amsterdam
Data Scientist at OLX (Search and Recommendations)
carmine.paolino@olx.com
@paolino
http://paolino.me
3. 3
Hagop Boghazdeklian
MSc in DS and Econometrics from Aix-Marseille University
Data Scientist at OLX (Fraud, Trust and Safety)
hagop.boghazdeklian@olx.com
14. 14
Deep Neural Networks
Overview of approaches
❖ Basic and pragmatic approach
❖ Two-steps classification
➢ Object detection (bounding-boxes)
➢ Classification of cropped images
❖ Increase of robustness in recent CNN architectures
❖ Transfer Learning, Data Augmentation and Fine-Tuning
15. 15
Deep Neural Networks
Overview of approaches. Valev, Schuman and Sommer
A Systematic Evaluation of Recent Deep Learning Architectures for Fine-Grained Vehicle Classification (2018)
17. 17
Example of pose-normalized part-CNNs
Figure 3: Overview of the part-CNN pipeline on pose normalized parts
part-CNN
Overview of approaches. Source: Branson, Van Horn, Belogie, Peronize.
Bird species categorization using prose normalized deep convolutional nets (2014)
18. 18
Ensemble of networks
Overview of approaches. Lin, RoyChowdhury, Maji. Bilinear CNN models for fine-grained classification (2015)
19. 19
Attention mechanism
Overview of approaches. Zheng Fu, Luo. Learning Multi-Attention Convolutional Neural Network for Fine-Grained Image Recognition (2017)
25. 25
Classical approaches
State-of-the-art: H. Zheng, J. Fu, T. Mei, and J. Luo, “Learning Multi-attention Convolutional Neural Network for Fine-Grained Image Recognition,” Proc. IEEE
Int. Conf. Comput. Vis., vol. 2017–October, pp. 5219–5227, 2017.
Part-detection Templates Pose alignment Classification
y
26. 26
Multi-attention
State-of-the-art: H. Zheng, J. Fu, T. Mei, and J. Luo, “Learning Multi-attention Convolutional Neural Network for Fine-Grained Image Recognition,” Proc. IEEE
Int. Conf. Comput. Vis., vol. 2017–October, pp. 5219–5227, 2017.
Highest activations
VGG-19
Grouping
Candidate part attentions
27. 27
Final classification
State-of-the-art: H. Zheng, J. Fu, T. Mei, and J. Luo, “Learning Multi-attention Convolutional Neural Network for Fine-Grained Image Recognition,” Proc. IEEE
Int. Conf. Comput. Vis., vol. 2017–October, pp. 5219–5227, 2017.
28. 28
Part-CNN for final classification
Figure 3: Overview of the part-CNN pipeline on pose normalized parts
State-of-the-art. Source: Branson, Van Horn, Belogie, Peronize.
Bird species categorization using prose normalized deep convolutional nets (2014)
Region zooming
29. 29
Joint learning
State-of-the-art: H. Zheng, J. Fu, T. Mei, and J. Luo, “Learning Multi-attention Convolutional Neural Network for Fine-Grained Image Recognition,” Proc. IEEE
Int. Conf. Comput. Vis., vol. 2017–October, pp. 5219–5227, 2017.
30. Thank you!
H. Zheng, J. Fu, T. Mei, and J. Luo, “Learning
Multi-attention Convolutional Neural Network for
Fine-Grained Image Recognition,” Proc. IEEE Int. Conf.
Comput. Vis., vol. 2017–October, pp. 5219–5227, 2017.
B. Zhao, J. Feng, X. Wu, and S. Yan, “A survey on deep
learning-based fine-grained object classification and
semantic segmentation,” Int. J. Autom. Comput., vol.
14, no. 2, pp. 119–135, 2017.