MODEL_ZOO.md

DDN Model Zoo

• For all the three Datasets we provide the jointly trained models (CNN and DN). These models are present in the directory - "models".
• We also used pre-trained models (both CNN and DN) to make the end to end learning faster. These models are located in folder named "pre-trained-models".
  1. Charades
     • For Charades we used the SlowFast model which can be downloaded from the PySlowFast repository.
     • We used the SlowFast R50 model trained on Charades using the standard training protocol. We updated the config file to accomodate DDNs.
     • The config files are present in a folder called configs inside the Code directory. Please select the config which corresponds to the DN model you are using.
  2. TaCOS and Wetlab
     • The scripts to train the CNN model were adapted from this repository. We modified the code to accomodate the DDN and joint training.
     • For TaCOS and Wetlab we used a pre-trained InceptionV3 model.
     • Model and usage details are given in the repository we mentioned above.
     • The config files are present in a folder called configs inside the corresponding code directory. Please select the config which corresponds to the DN model you are using.