Class Activation Map

Unofficial Pytorch Implementation of 'Learning Deep Features for Discriminative Localization'

Reference: Learning Deep Features for Discriminative Localization, CVPR2016

I used the Networks that trained ImageNet data from torchvision.models.

Requirements

• torch (version: 1.2.0)
• torchvision (version: 0.4.0)
• Pillow (version: 6.1.0)
• matplotlib (version: 3.1.1)
• numpy (version: 1.16.5)

Usage

Arguments

• --gpu-no: Number of gpu device (-1: cpu, 0~n: gpu)
• --network: Network for backbone (Possible networks: resnet50, resnext50_32x4d, wide_resnet50_2, googlenet, densenet161, inception_v3, shufflenet_v2_x1_0, mobilenet_v2, mnasnet1_0)
• --image: Input image path
• --topk: Create k Class Activation Maps (CAMs) with the highest probability
• --imsize: Size to resize image (maintaining aspect ratio)
• --cropsize: Size to crop cetenr region
• --blend-alpha: Interpolation factor to overlay the input with CAM
• --save-path: Path to save outputs

Example Script

python cam.py --image imgs/input/img1.jpg --topk 3 --imsize 256 --network resnet50

Results