Video Outpainting Localizer

04/2023 - 09/2023

Project description

Deep learning model for video outpainting localization. The model is able to localize the outpainted region in a video frame and to generate a binary mask that highlights the outpainted region. The implementation is based on the RAFT model.

The main goal of the project was to modify the RAFT model and in particular we focused on data augmentation and loss function. The model has achieved a F1 score of 0.786 on the test set.

Code not available due to non-disclosure agreement.

Artifacts

Video

Team and role

Team size: 2

  • The project was carried with a pair programming approach.
  • I was responsible for the implementation of the data augmentation and the loss function.

Tech stack

Python NumPy