This repository includes the codes for the study of CAM models, such model can visualize regions of input image that activates the decision of the model, which is typically used to interpret the inner working of an image classifier.
This program is tested on Python 3.9.0, but should work on later version of python. It is also recommended to run this in a device with GPU as it is not tested on a CPU-only machine.
If you do not have virtual environment module, install it by
pip install virtualenv
Then create a virtual environment by
virtualenv venv
A folder ./venv/ should appear. Then, activate venv by (assuming you're using
Windows 10)
./venv/Scripts/activate
Install torch and torchvision (with GPU recommended)
from the official pytorch website.
GPU example:
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu117
Install requirements by running
pip install -r requirements.txt
(Must do)
Run python prepare.py to download dataset, splits and models.
Decompress pretrained models from ./models.rar so that /model/
appear.
Perform data augmentation by running python augment.py.
(Optional)
To train a model, use main.py (need to understand flags)
To inference, use inference.py (need to understand flags)
To evaluate (accuracies), use evaluate.py (need to understand flags)
