Run in Streaming Mode

[arshdeep-liat-ai]

Is there a way to run this on a video in streaming manner rather than processing the whole video at once?

[pranay-junare]

+1
I am also interested in this. Thanks.

[sourav-bz]

yes, online mode would help immensely

[chenweisomebody126]

@arshdeep-liat-ai @pranay-junare @sourav-bz
I've implemented exactly this — a streaming mode where you process one frame at a time with constant memory usage.

The pipeline works as: provide a point prompt on frame 1, then feed subsequent frames one by one. A ring-buffer memory bank (7 frames) keeps memory bounded, so it works on arbitrarily long videos without loading anything upfront.

It's an iOS-native implementation (CoreML + Metal), designed for real-time camera feeds:

• GitHub: https://github.com/chenweisomebody126/SegmentKit
• Docs: https://segmentkit.dev

The key API is just startTracking(frame:, point:) for the first frame, then track(frame:) for every frame after that.