SiamMask

Why we included this project

For teams working on video object tracking and segmentation, SiamMask is the official reference implementation of a method that merges both tasks into a single forward pass. Instead of choosing between a bounding box and a pixel mask, the tracker outputs both at once, which matters for applications like video object segmentation, fine-grained motion analysis, and per-frame annotation of clips. The repo includes trained models ready to run plus full training code, so you can reproduce the published VOT and DAVIS numbers or fine-tune on your own footage. It is built on PyTorch and clearly documented, making it a solid baseline for building or extending a real-time tracking and segmentation pipeline, though it stays research-oriented rather than a plug-and-play product.