Skip to content
A PyTorch implementation of WS-DAN (Weakly Supervised Data Augmentation Network) for FGVC (Fine-Grained Visual Classification)
Branch: master
Clone or download
Latest commit def8d80 May 27, 2019
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
images first pushing May 7, 2019
models first pushing May 7, 2019
.gitignore first commit: add .gitignore May 7, 2019
LICENSE Create LICENSE May 27, 2019
README.md modified README May 8, 2019
dataset.py first pushing May 7, 2019
train_wsdan.py first pushing May 7, 2019
utils.py first pushing May 7, 2019

README.md

WS-DAN.PyTorch

A PyTorch implementation of WS-DAN (Weakly Supervised Data Augmentation Network) for FGVC (Fine-Grained Visual Classification). (Hu et al., "See Better Before Looking Closer: Weakly Supervised Data Augmentation Network for Fine-Grained Visual Classification", arXiv:1901.09891)

NOTICE: This is NOT an official implementation by authors of WS-DAN.

Attention Cropping and Attention Dropping

Fig1

The framework introduce an attention based method for extracting more detailed features and more object's parts by Attention Cropping and Attention Dropping, see Fig 1.

Training Process and Testing Process

Fig2a

Fig2b

Bilinear Attention Pooling (BAP)

Fig3

Usage

This code repo contains WS-DAN with feature extractors including VGG19, ResNet(34, 50, 101, 152), and Inception_v3 in PyTorch form. The default feature extractor is Inception_v3, and this can be modified conveniently in train_wsdan.py:

# feature_net = vgg19_bn(pretrained=True)
# feature_net = resnet101(pretrained=True)
feature_net = inception_v3(pretrained=True)

net = WSDAN(num_classes=num_classes, M=num_attentions, net=feature_net)
  1. git clone this repo.
  2. Prepare image data and rewrite dataset.py for your CustomDataset.
  3. $ nohup python3 train_wsdan.py -j <num_workers> -b <batch_size> --sd <save_ckpt_directory> (etc.) 1>log.txt 2>&1 & (see train_wsdan.py for more training options)
  4. $ tail -f log.txt for logging information.
You can’t perform that action at this time.