Attributional-Robustness-Training

Training methodology for attributional robustness and its application in weakly supervised object localization

Code for paper: Attributional Robustness Training using Input-Gradient Spatial Alignment (ECCV 2020)

Requirements

• Python 3.6
• Pytorch (≥ 1.1)
• Python bindings for OpenCV.
• TensorboardX
• OpenCV-python

Running code on CIFAR-10

---

cd cifar
python cifar_robust_train.py 

Evaluation

python eval_model.py 

Running code on CUB-200 for WSOL

---

cd WSOL_CUB
bash scripts/prepare_dataset.sh
bash scripts/run_resnet_beta.sh

Evaluation

bash scripts/eval_resnet_beta.sh

Pretrained models on CIFAR-10 and CUB-200

CIFAR-10: https://drive.google.com/file/d/1Xjn3kX_Lh887eIKicWhZFBtgRgKpCg6q/view?usp=sharing

CUB-200: https://drive.google.com/file/d/1LMUDHh6deCQ54mpVNqXQnYXXuIiAruzv/view?usp=sharing