VisualBackProp - visualization method for convolutional neural networks
Detailed description of the VisualBackProp can be found in: https://arxiv.org/abs/1611.05418
Content:
- inputImages - folder with input images
- outputImages - folder for output images
- vis.lua - implementation of VisualBackProp
- run.lua - loads images, runs visualization method, saves images
- model.t7b - model trained to predict steering wheel angle
mkdir outputImages
qlua run.lua
Script generates set of 3 images for each input image. Each set contain:
- Input image with visualization mask overlaid in red
- Averaged feature maps
- Intermediate masks