Segment Image (HuggingFace)

How it fits

Typical backends

Caveats

• I/O contractThe output arm is a single downstream-resolved choice fixed at deploy time; every arm reprojects the same segments, so masks, boxes and classes stay aligned.
• CompatibilityThe backend is chosen from the repo contents named by model, so the repo must ship a segmentation model in one of the supported formats; a weights file is preferred over an ONNX export when both are present.
• Latencymodel is hot-swappable at runtime: a change fetches the new snapshot and frees the previous GPU allocation in place, so the first frame after a swap pays the download and load cost.
• Parameter interactionthreshold is the base confidence and class_thresholds lowers or raises it per class; candidates are fetched at the minimum needed threshold and re-filtered per class.
• Parameter interactionobject_classes empty keeps every class, max_classes caps the number of distinct classes kept, max_segments caps the segment count, and sort_by_area orders segments largest-first.
• CompatibilityWeights download from the Hub on first use, so a cold container needs outbound network unless the cache is pre-seeded.