Demo (click on GIF for the full video) :
Wished for a dropbox which works with videos! Well wait no more!! Search moments that you relish, you want to share!
The app looks through a library of videos, samples frames every few seconds, and generates natural language captions using a CNN+LSTM for each video frame, written in PyTorch. Next, all the captions are indexed using lucene and users can search this index using a web interface.
We will explain in the presentation
Setting up the infrastructure for video feature extraction. Setting up the indexing and searching aspects.
Bunch of technologies like Lucene, PyTorch, LSTM and CNN.
Scaling to 1000s of videos database