DeepLearning-Big-Project

50.039 Deep Learning, Y2021
Group Member: Zachary, Leng Sicong, Li Jiaxi

Dataset and pre-trained models preparation

Please download dataset and pre-trained models zip file from Google Drive
Then unzip the file under the root directory:

unzip Data_and_Model.zip

Training and Testing

Charades-STA dataset

To train the model from scratch, run the following commands:

python main_predictor.py --visual_dim 1024 --model_dir saved_char/ --task charades  --pos_freeze --vq_mi --vq_mi_lambda 0.3 --vv_mi --vv_mi_lambda 0.3 --mode train

To test existing pre-trained model and reproduce the result:

python main_predictor.py --visual_dim 1024 --model_dir saved_char/ --task charades  --pos_freeze --vq_mi --vq_mi_lambda 0.3 --vv_mi --vv_mi_lambda 0.3 --mode test
TACoS dataset

To train the model from scratch, run the following commands:

python main_predictor.py --model_dir saved_tacos/ --task tacos  --pos_freeze --vq_mi --vq_mi_lambda 0.3 --vv_mi --vv_mi_lambda 0.3 --mode train

To test existing pre-trained model and reproduce the result:

python main_predictor.py --model_dir saved_tacos/ --task tacos  --pos_freeze --vq_mi --vq_mi_lambda 0.3 --vv_mi --vv_mi_lambda 0.3 --mode test

Report

Please refer to PDF or Google Doc for more details.