A pytorch implementation of a text to videos GAN that makes use of MoCoGAN to create new text, of Caffe library with pretrained S2VT to get the description of videos and of UCF-101 Dataset to train models. An LSTM model is trained on the results of S2VT to classificate user input to classes of UCF_101
In this notebook the results can be seen. Here it is explained how to use the project and how to generate videos.
Here the transformations for videos in UCF_101 are shown with results. The statistics of the dataset are computed in the last cell.
Here the process of loading the LSTM model pretrained and the usage of it to classificate NL are shown.
To use notebooks, mocogan you can import the environment called mocogan.yml found into the directory called Environments.
To do so, simply use command conda env create -f Environments/mocogan.yml.
Same thing apply if you want to use the s2vt captioner: conda env create -f Environments/s2vt_captioner.yml.