Error Recognition

Please download or extract features for the data following the instructions in the Feature Extraction repository.

Once you have the features, you can run the code in this repository to train the models.

We have code three tasks:

1. Error Recognition
   • Multimodal Data
     • Video
     • Audio
     • Text
     • Depth
2. Early Error Recognition
   • Video Data

How to run the code

1. Download or extract features for the data.
2. Place the features in the respective folders.
3. Change the paths in the /core/config.py file.
4. Copy the respective script from scripts folder to the root directory.
5. Run the script.

How to evaluate the models

Our reported results correspond to the following threshold values:

1. Step Split: 0.6
2. Recordings Split: 0.4

Note the best epoch of the trained models and add it accordingly to the /core/evaluate.py file.

Use it to get the evaluation csvs for different thresholds.