This project has been developed for the Images and Videos Analysis class by t0re199 and adel3roman6.
The project required to perform a Multi-Class and Multi-Label Classification on an unbalanced dataset. In particulat, the dataset contains frames of film trailer (the number of frames can be different for each film).
In this project the well-known image classification architectures ResNet34 and Vgg-16 have been used. Since PyTorch has been used as Gradient Computing framework, the above cited architecture were taken from the TorchVision Library.
Since the dataset contains unlabeled data too, in order to improve the model's extracted features quality a self-supervised learning strategy has been applied during the last epochs of the training process.