Authors: Antonio Commisso, Francesco Pasceri
Basketball video actions dataset.
10 labels - Strongly unbalanced
Video augmentation and finetuning of state-of-the-art architectures trained on kinetics-400
We trained different models in order to achieve a fixed baseline (85% of avg precision).
The final version of the notebook is 'ProgettoAIV_finale.ipynb'. Within the file you will find the whole code to reproduce our results with the same network from scratch. Within the old_notebooks directory we moved all the other architectures.