sam3-api

Text prompts enable open-vocabulary segmentation with SAM3 by passing descriptive text to the API. The unified Python API processes these prompts to isolate target objects across static images and video streams without requiring bespoke tooling.

Geometric prompts like bounding boxes with positive or negative labels, alongside point prompts, drive SAM3 video tracking. You apply these click-based prompts to initial frames, and the API streams segmentation masks across subsequent video frames programmatically.

Batch inference is supported natively by the SAM3 Python API, allowing you to process multiple images simultaneously. This integrates with model-building guidance to enable fast experimentation and high-throughput promptable segmentation workflows.

PyTorch is required as the SAM3 API relies on PyTorch-based data structures and post-processing implementations. The API exposes Python interfaces to load the SAM3 model, set images, and apply prompts compatible with this framework.

SAM3 provides an integrated workflow that applies the same prompt types to both static images and streaming video. The API manages video tracking by propagating segmentation masks across frames, maintaining object continuity throughout the stream.

A unified API eliminates the need for bespoke tooling by exposing text, box, and point prompts through a single Python interface. This streamlines model building and ensures consistent post-processing across both image and video segmentation tasks.