SmoothGrad with PyTorch
PyTorch implementation of SmoothGrad [1].
WIP, not tested on GPU.
Dependencies
Python 2.7
PyTorch
torchvision
tqdm
Examples
python main.py --image samples/cat_dog.png [--no-cuda] [--guided]
With the --guided
option, you can generate smoothed maps from guided backproped gradients.
Model: ResNet-152 pre-trained on ImageNet
Prediction: bull mastiff - 54.3% @1
#samples: 100
References
[1] D. Smikov, N. Thorat, B. Kim, F. Viégas, M. Wattenberg. "SmoothGrad: removing noise by adding noise". arXiv, 2017