wufan-tb / cam_wsol

weakly supervised object localization based on cam

A Toy Weakly Supervised Object Localization using CAM

a toy weakly supervised object localization using cam(Class Activation Mapping, proposed in "Learning Deep Features for Discriminative Localization").

Steps:

1. download classification dataset(dog vs cat) from kaggle

2. unzip dataset to "data/dog_vs_cat":

   after unzip, folder structure should be:

   data
   └── dog_vs_cat
      ├── test
      │   ├── 1.jpg
      │   └── ...
      └── train
         ├── cat.0.jpg
         └── ...
   

3. traing

   python train.py --data_path data/dog_vs_cat
   

4. inference

   python inference.py --input {single image path or images folder}
   

Demo: