Description

Adds a new SAM3 workflow block (v3) that supports RLE mask output format

• New SAM3 v3 block (roboflow_core/sam3@v3): Supports both rle and polygons output formats via the output_format parameter
• New RLE_INSTANCE_SEGMENTATION_PREDICTION_KIND: A new Kind type for RLE-based instance segmentation predictions. RLE data is stored in sv.Detections.data["rle_mask"] and decoded numpy masks are populated in sv.Detections.mask
• RLE serialization/deserialization: Added serialise_rle_sv_detections() and deserialize_rle_detections_kind() functions to handle RLE mask data
• New InstanceSegmentationRLEPrediction model: Pydantic model for RLE-based predictions with rle field instead of points.
• Visualization blocks updated: Halo, Mask, Polygon, Icon, and Model Comparison visualizations now accept RLE_INSTANCE_SEGMENTATION_PREDICTION_KIND as input

Type of change

• New feature (non-breaking change which adds functionality)

How has this change been tested, please provide a testcase or example of how you tested the change?

• Locally, using the workflow: https://app.roboflow.one/leandro-starter/workflows/edit/sam3-v3-vs-v2 (running inference with the new changes locally)