CAM

Implementation of Learning Deep Features for Discriminative Localization(arxiv)

Dependency

Need opencv-python(cv2) and pytorch

pip install opencv-python

Training

To train a model

python train.py --dataset CIFAR --dataset_path ./data --model_path ./model --model_name model.pth --img_size 128 --batch_size 32 --epoch 30 --log_step 10 --lr 0.001

Arguments

• --dataset : Specify which dataset you use. Three types are supported: (STL, CIFAR, OWN) If you want to train model with your own dataset, use OWN

• --dataset_path : Specify the path to your dataset. If you use STL10 or CIFAR10, it will download the dataset at the path.

• --model_path : Specify the path where the model to be saved

• --model_name : Specify the name of .pth file

• --img_size : The size of images to train

• --batch_size : The number of images in each batch

• --epoch : The number of epochs to train

• --lr : Learning rate

• --log_step : The number of iterations to print loss

• -s, --save_model_in_epoch : Basically the model will be saved after an epoch finished. If -s is true, the model will be saved after each log_step too.

Create CAM

it needs saved model

python create_cam.py --dataset CIFAR --dataset_path ./data --model_path ./model --model_name model.pth --result_path ./result --img_size 128 --num_result 1

Arguments

• --dataset : Specify which dataset you use, (STL, CIFAR, OWN) If you want to test model with your own dataset, use OWN

• --dataset_path : Specify the path to your dataset. If you use STL10 or CIFAR10, it will download the dataset at the path.

• --model_path : Specify the path where the model is saved

• --model_name : Specify the name of .pth file

• --result_path : Specify the path where the CAM and the original image to be saved

• --img_size : The size of images to save

• --num_result : The number of result to create. It will be randomly chosen from the test dataset.

Reference

Codes of create_cam.py is influenced by https://github.com/metalbubble/CAM