Implemented a video summarization system. The system takes a video as an input and outputs unique frames along with their captions.
Note: This project is still work-in-progress. Currently, the frontend uses BEAM search by default with width 3. For now, you can change it at /frontent/src/components/VidCap/VidCap.js
.
This repository is divided into three components. The frontend, which is created using ReactJS, the backend, made with flask which servers our Tensorflow model, and the training notebook. You don't have to retrain, you can use my trained checkpoints instead: https://drive.google.com/drive/folders/1aUEPjykLXqI_K70o09fFjreOkSCF8H_l?usp=sharing
Put the checkpoint files at backend/ckpt/ckpt6/
. You can change the directory and checkpoint in backend/utils.py
Change UPLOAD_FOLDER
to your desired location in backend/server.py
- Start the backend.
cd backend
python server.py
- Start the frontend.
cd frontend
npm i
npm start