This is the internal beta implementation of our Action-Verificaiton project.
Our environment:
Ubuntu 18.04.5 LTS
CUDA 10.2.89
Pytorch 1.7.1
To install other packages:
$ pip install -r requirements.txt
The detailed setup of model, training, dataset are configured in [config file], dir configs/ contains some examples. The following are main commands used to play with this project.
For training:
$ CUDA_VISIBLE_DEVICES=[X,X,X,X] python train.py --config [train config]
For testing:
Mode 1: Evaluate all models in [root_path/saved_models] and save log file to [root_path/logs].
$ CUDA_VISIBLE_DEVICES=[X,X,X,X] python eval.py --config [eval config] --root_path [root_path]
Mode 2: Evaluate one model in [model_path] and save log file to [temp_log] if not especially speicify.
$ CUDA_VISIBLE_DEVICES=[X,X,X,X] python eval.py --config [eval config] --model_path [model_path]
First follow the instructions in here and then run the following command:
$ CUDA_VISIBLE_DEVICES=[X,X,X,X] python vis.py --config [vis config] --model_path [model_path]