facebookresearch/co-tracker
tracks any pixel in video with transformer-based model, supports dense and long-term point tracking even through occlusion
View on index · View in 3D Map
Related repos
- Video models like SlowFast help computers spot actions and understand what’s happening in videos by combining slow and fast motion cues
- video and image watermarking code for training and inference
- Reconstructs 3D scenes from 1000+ images in one forward pass
- EdgeTAM: on-device SAM 2 variant for segmentation and tracking in videos, runs faster, processes 16 FPS on iPhone 15 Pro Max
- Library for video understanding research
- VGGT is a neural network that predicts 3D scene attributes like camera parameters, depth, and point maps from images
- Tracks 3D points across multiple camera views
- data augmentations for audio, image, text, and video
- State-of-the-art object detection and segmentation library from Facebook AI Research
- Adds tracking capabilities to any object detector
