Show HN: Open-source, real-time object tracking for live video feeds
github.comtamserve runs Meta's SAM 3 on live camera feeds. SAM 3's own tracker handles one recorded video at a time, runs at about 8 fps, and runs out of memory on long videos. tamserve uses SAM 3 to find objects every few frames and a lightweight tracker (EdgeTAM) to follow them in between, with one copy of each model shared across all your cameras.
*On one RTX 4090:*
* *Any object you name:* 4x faster than SAM 3's tracker, at 92% of its accuracy * *People:* 16 cameras at once, 14x less GPU per frame, and slightly more accurate than SAM 3 (89.6 vs 87.8 HOTA) * *Memory:* flat at 3.5 GB on a 3,000 frame video, where SAM 3 runs out at frame 423
*Worth a look if you're building:*
* Warehouse or factory safety (forklifts, pallets, hard hats) * Security or retail camera analytics * Home or elder care monitoring * Robotics, wildlife or lab monitoring * Annotation tools for long videos
`pip install "tamserve[sam3]"`
Please give me any and all feedback or edit tamserve to meet your needs, and a star helps if you find it useful!
We’re about to launch a new real-time API for SAM3.1, that’s optimized for low latency, if folks are interested.
We’re seeing up to 3.5x speedup (lossless) over the native PyTorch implementation. ~40ms (ours) vs 150ms (eager-mode).
Second, I have a meta question (apologies if this is not the right place to ask): this has been posted twice in the last few days (https://news.ycombinator.com/item?id=50488117), the first time getting only 1 point, this time 81 so far.
Is it ok by hn rules and etiquette to post the same project repeatedly within days? Some projects I posted in the past had a good response in comments but not enough points to stay around, and would like to repost them if possible so they reach more people (not looking for customers on those projects, just feedback and PRs).