amazon-science/siam-mot
A region-based Siamese neural network for real-time multi-object tracking in video at 17 FPS on GPU.
Not currently ranked — collecting fresh signals.
star history
SiamMOT is a region-based Siamese multi-object tracking network that simultaneously detects and associates object instances across video frames. It includes a motion model that estimates instance movement between frames to enable association of detections. The project provides two tracker variants—one with implicit motion modeling and one with explicit motion modeling—and experiments on MOT17, TAO-person, and Caltech Roadside Pedestrians datasets. Inference runs at 17 FPS for 720P video on a single modern GPU.
Frequently asked
- What is amazon-science/siam-mot?
- A region-based Siamese neural network for real-time multi-object tracking in video at 17 FPS on GPU.
- Is siam-mot open source?
- Yes — amazon-science/siam-mot is open source, released under the Apache-2.0 license.
- What language is siam-mot written in?
- amazon-science/siam-mot is primarily written in Python.
- How popular is siam-mot?
- amazon-science/siam-mot has 500 stars on GitHub.
- Where can I find siam-mot?
- amazon-science/siam-mot is on GitHub at https://github.com/amazon-science/siam-mot.