This is the merge of two separate projects regarding egocentric action recognition with different goals. Internally each of them has a README.
- EGO_Project_Group4 : Exploration of convolutional neural network trained end-to-end from raw data for predicting the action. In particular, I3D and TSM.
- AggregationMethods : Exploration of temporal aggregation of the features extracted by the convolutional neural networks using average poolinand a temporal relation network TRN.
- MyTA3N : Exploration of the domain adaptation task by using adversarial classifiers and its extension by introducing transformers.
For running any experiment it is necessary to have a python environment properly configured.
$python pip install -r requirements.txt
The dataset it can be downloaded directly from the EPIC-KITCHENS website. Instructions can be found inside the EGO_Project_Group4 folder. Features to be used with other networks can be extracted from them.