search

hulianyuyy / DSTA-SLR

Dynamic Spatial-Temporal Aggregation for Skeleton-Aware Sign Language Recognition (COLING2024)
Other
6 stars 2 forks source link

fix errors and provide installation instruction #1

Closed tanthinhdt closed 4 months ago

tanthinhdt commented 4 months ago

Hi, I found that you haven't provided any installation instructions or requirements yet, so I have added them in this pull request. Moreover, when running your code, I encountered some errors in main.py and feeder.py and corrected them for you.

hulianyuyy commented 4 months ago

Many thanks for your code checking! When i ran the code, it doesn't require cuda 12.1. I run the code with cuda 11.3. Which package requires cuda==12.1 ?

tanthinhdt commented 4 months ago

I used to have CUDA 10.1. When I installed torch-geometric, it removed the current version of torch I installed for CUDA 10.1 and installed the latest one (2.2.2). Then, when installing torch-scatter, I encountered an error that said the current CUDA version (10.1) didn't match with that used to compile torch (12.1). Only when I change CUDA version to 12.1, everything works out smoothly.

hulianyuyy commented 4 months ago

Many thanks for your response. I suspect that the lastest pytorch requires cuda==12.1. This repo acturally is compatible with cuda<12.0. I will merge the changes.