Deep-Flow-Guided-Video-Inpainting

Why we included this project

The key idea here is that a video isn't just a stack of independent frames. This CVPR 2019 method estimates optical flow across frames and uses that motion to fill missing regions, so the repaired areas stay consistent instead of drifting between frames the way per-frame inpainting tends to. If you need to erase unwanted objects from footage, whether for post-production cleanup or preparing clean training data, the repo ships inference scripts and pre-trained weights that get you to working results without rebuilding the pipeline from scratch. Just note the code was built against PyTorch 0.4.0, so expect some dependency wrangling on a modern setup; the bundled demo frames, masks, and fixed-region weights give you a concrete way to verify it runs.