Curious about where is the ResNet looking at? This repo visualizes the class specific saliency map or discriminative location. It is Torch re-implementation of Learning Deep Features for Discriminative Localization, Bolei Zhou et. al., without modifying the network or re-training. (MatConvNet re-implementation can be found at this repo.) We look directly through the ResNet to see the world.
- The code relies on Torch, which should be downloaded and built before running the experiments. Download the code:
git clone https://github.com/zhanghang1989/visualResNet_torch.git
Download the pre-trained models from Facebook ResNet repo
Run the progrem:
th visual.lua resnet-50.t7 data/1.JPEG data/2.JPEG
- Visualize all the images in a folder
th visual.lua resnet-50.t7 data/*