Question about Real Time Video Segmentation Performance Benchmark

Hi Team,

want to know how to evaluate the real time performance of sam3 model predictor over video?
Any scripts or notebook support fps benchmark?

In the paper, here are some related inforamtion:

> To support real-time inference (30 FPS) on videos in practical applications (e.g., a web demo), we parallelize the inference over multiple GPUs: up to 10 objects on 2 H200s, up to 28 objects on 4 H200s, and up to 64 objects on 8 H200s.

Also in Table 7:

> In video, the inference latency scales with the number of objects, sustaining near real-time performance for ∼ 5 concurrent objects.

Want to know details about how to `test` this benchmark. Any concise benchmark script or a short decision matrix showing expected fps vs. GPU class and tracked object count (with conservative estimated bands).

Thank you

Start with the repository's example notebooks and the SAM 3 video predictor entry point, then determine how inference latency and FPS should be measured across tracked object counts and GPU configurations. Done means documenting a reproducible benchmark procedure, or adding the concise benchmark script or decision matrix requested in the issue.

Written by the indexing model from the issue text.