In this repository are saved the main artifacts of my bachelor thesis. The title of the thesis is "Heterogenous Graph Neural Networks for Action Prediction on Egocentric Action Scene Graphs". As the title says, the main focus of this work was to experiment with Heterogenous GNN, a particular type of GNNs capable of handling heterogenous graphs, to perform action prediction on EASG, a novel form of representation of egocentric videos.
The three documents present in this repository are:
- The actual thesis in pdf.
- The slides used to present this work to the University Commition.
- The notebook containing the code used to perform such experiments.
The notebook is also accompanied by the annotation files used. The files containing the additional features are available in the EASG main repo.