Deep Learning for Computer Vision: Data Augmentation (UPC 2016)
4 likes4,103 views
The document discusses the use of data augmentation in deep convolutional neural networks for image classification, specifically referencing the ImageNet challenge. It outlines techniques to reduce overfitting, such as reducing network capacity, dropout, and various data augmentation strategies that enhance training by altering input images. The document also highlights the importance of creating robust features through these augmentations and their effectiveness in unsupervised learning scenarios.
![[course site]
Augmentation
Day 2 Lecture 2
Eva Mohedano](https://image.slidesharecdn.com/dlcvd2l2augmentation-160802094637/85/Deep-Learning-for-Computer-Vision-Data-Augmentation-UPC-2016-1-320.jpg)
![Augmentation for discriminative unsupervised
feature learning
Select random location k and crop 32x32 window
(restrictions: region must contain objects or part of the
object: high amount of gradients)
Apply a transformation [translation, rotation, scalig, RGB
modification, contrast modification]
...
Generate augmented dataset: 16000 classes of 150 examples each
Class k=1, with 150 examples
11](https://image.slidesharecdn.com/dlcvd2l2augmentation-160802094637/85/Deep-Learning-for-Computer-Vision-Data-Augmentation-UPC-2016-11-320.jpg)
Deep Learning for Computer Vision: Data Augmentation (UPC 2016)
- 1. [course site] Augmentation Day 2 Lecture 2 Eva Mohedano
- 2. Introduction ImageNet Classification with Deep Convolutional Neural Networks, Krizhevsky A., 2012 ImageNet Large-Scale Visual Recognition Challenge (ILSVRC) 1.2 million training images, 50,000 validation images, and 150,000 testing images Architecture of 5 convolutional + 3 fully connected = 60 million parameters ~ 650.000 neurons. Overfitting!! 2
- 3. ● Reduce network capacity ● Dropout ● Data augmentation Ways to reduce overfitting 3
- 4. ● Reduce network capacity ● Dropout ● Data augmentation Ways to reduce overfitting 1% of total parameters (884K). Decrease in performance 4
- 5. ● Reduce network capacity ● Dropout ● Data augmentation Ways to reduce overfitting 37M, 16M, 4M parametes!! (fc6,fc7,fc8) 5
- 6. Ways to reduce overfitting ● Reduce network capacity ● Dropout ● Data augmentation Every forward pass, network slightly different. Reduce co-adaptation between neurons More robust features More interations for convergence 6
- 7. Ways to reduce overfitting ● Reduce network capacity ● Dropout ● Data augmentation 7
- 8. Data Augmentation During training, alterate the input image (Krizhevsky A., 2012) - Random crops on the original image - Translations - Horitzontal reflections - Increases size of training x2048 - On-the-fly augmentation During testing - Average prediction of image augmented by the four corner patches and the center patch + flipped image. (10 augmentations of the image) 8
- 9. Data Augmentation Alternate intensities RGB channels intensities PCA on the set of RGB pixel throughout the ImageNet training set. To each training image, add multiples of the found principal components Object identity should be invariant to changes of illumination 9
- 10. Augmentation for discriminative unsupervised feature learning Discriminative Unsupervised Feature Learning with Exemplar Convolutional Neural Networks, Dosovitskiy, A., 2014 MOTIVATION ● Large datasets of training data ● Local descriptors should be invariant transformations (rotation, translation, scale, etc) WHAT THEY DO ● Training a CNN to generate local representation by optimising a surrogate classification task ● Task does NOT require labeled data 10
- 11. Augmentation for discriminative unsupervised feature learning Select random location k and crop 32x32 window (restrictions: region must contain objects or part of the object: high amount of gradients) Apply a transformation [translation, rotation, scalig, RGB modification, contrast modification] ... Generate augmented dataset: 16000 classes of 150 examples each Class k=1, with 150 examples 11
- 12. Augmentation for discriminative unsupervised feature learning Generate augmented dataset: 16000 classes of 150 examples each Example of classes Example of examples for one class 12
- 13. Augmentation for discriminative unsupervised feature learning Classification accuracies Superior performance to SIFT for image matching. 13
- 14. Summary Augmentation helps to prevent overfitting It makes network invariant to certain transformations: translations, flip, etc Can be done on-the-fly Can be used to learn image representations when no label datasets are available. 14