The PyTorch code of our paper titled Convolutional STN for Weakly Supervised Object Localization
The experiments are performed on two separate datasets, CUB-200 and ImageNet-1K. There are single scale and multi-scale models(with Feature Pyramid Network).
To run the code for trainining on CUB-200 dataset
python cub_train_multiscale.py -e 20 -bs 16
For training on ImageNet
python imagenet_train_multiscale.py -e 20 -bs 16
Follow the similar commands for training single scale models with files cub_train.py
and imagenet_train.py
.
The paper is available in Arxiv.