Combinational Class Activation Maps for Weakly Supervised Object Localization (WACV 2020)

Combinational Class Activation Maps for Weakly Supervised Object Localization

Environments

python == 3.6

pytorch == 0.4.0

opencv == 3.4.2

Dataset

CUB-200-2011

PATH: ./CUB_200_2011

Pre-trained model

VGG16 pretrained on ImageNet

NL-CCAM for CUB-200-2011

Training and Test for your own model

Change codes for your model

Training: train_cub.py

1. model
2. dataloader

Evaluating: IoU_check_cub_5.py

1. load model
2. choose function for CCAM

Get the paper results

For getting bounding boxes, we should use a threshold value.

I fix the threshold value for quadratic function.

You can change threshold values for the best results of each combinational function.