What Do Deep Nets Learn? Class-wise Patterns Revealed in the Input Space

Introduction:

• Environment: Python3.6.5, PyTorch1.5.0
• Dataset: CIFAR-10, ImageNet-1k

Usage:

This is our paper link. You can firstly run select_canvas.py to choose a canvas. Then you can run generateMap.py to generate the mask and generatePos.py for clip. Finally you can extract the pattern by extractPattern.py. test.py is used for calculating the predictive power.

Notes

• Models used in our experiments are trained without normalization (i.e. torchvision.transforms.Normalization). To achieve this, for CIFAR-10 we just train from scratch and for ImageNet-1k, we fine-tune on the normalized trained model.
• We reconstruct CIFAR-10 for convenience by generateSet.py.