TimeT_clean

This is a better implementation of TimeT model.

Running the Code

1. Main training script:

   python exp_time_tuning_v2.py
   

   You can customize the run with command-line arguments:

   python exp_time_tuning_v2.py --device cuda:0 --batch_size 32 --num_epochs 100
   

2. For evaluation:

   python evaluation.py
   

3. To test the feature forwarder:

   python test_feature_forwarder.py
   

Dataset Files

The train.txt and val.txt files contain lists of video identifiers for the training and validation sets respectively. These files are used to split the dataset into training and validation subsets.

Logging and Visualization

The code uses Weights & Biases (wandb) for logging and visualization. Ensure you're logged in to your wandb account.

Performance curves

You should see something like this:

Where the evaluation is done on PascalVOC.