[judlaw]

Hi HN, we built this because running SAM3 on long videos without a GPU was painful due to memory constraints.

The core idea is automatic overlapping chunking based on available RAM, with IoU-based object ID remapping so tracking remains consistent across chunks.

It supports:

- Text/point/box/mask prompts - Video + image inference - Segment-level processing - Built-in profiling

Would love feedback -- especially from people running segmentation workloads on constrained hardware.

Thanks! Prashant & Ankuj