search

Henrymachiyu / ProtoViT

This code implements ProtoViT, a novel approach that combines Vision Transformers with prototype-based learning to create interpretable image classification models. Our implementation provides both high accuracy and explainability through learned prototypes.
MIT License
1 stars 1 forks source link

can't install the environment #1

Open hbaniecki opened 17 hours ago

hbaniecki commented 17 hours ago

Hi, neither pip install -r requirements.txt (error: file uses = instead of ==) nor conda create --name test --file requirements.txt (error: wrong pypi_0 channels) work for me to install the environment.

Henrymachiyu commented 8 hours ago

To reproduce the environment, only the packages specified in the README (such as timm, pytorch, augmentor, and cv2) are necessary. If you encounter issues with the environment setup, ensure that the PyTorch version is compatible with your system. Note that CUDA-related packages may also cause installation conflicts. Please use the requirements.txt file only as a reference, and install the specified packages manually to build the environment. Based on our test runs, the reproducibility of our work should be independent to pytorch and cuda versions. It only depends on the version of timm, since different version of timm implements ViT in a slightly different way.

I will leave this issue always open, in case someone has the same problem.