SAM3 Object Detection and Segmentation Scripts

Scripts for applying Meta's SAM3 model to detect and segment objects in images using text prompts. The repository contains Python scripts for generating bounding boxes and pixel-level masks on datasets from HuggingFace. It was created by uv-scripts and last updated in March 2026.

Use Cases

• Generate bounding box coordinates and category labels in an 'objects' column using the detect-objects.py script.
• Create pixel-level segmentation masks for detected objects using the segment-objects.py script.
• Perform zero-shot object detection on HuggingFace datasets by providing natural language descriptions as prompts.
• Extract per-instance segmentation masks for detailed object analysis from image collections.

Strengths

• Provides scripts for both bounding box detection and pixel-level segmentation.
• Enables zero-shot detection using natural language prompts without model retraining.
• Last updated in March 2026, indicating recent maintenance.

Limitations

• Unknown dataset size, row count, and specific image sources limit reproducibility assessment.
• No information on script performance metrics, accuracy, or computational requirements.
• Dependent on external HuggingFace datasets and the SAM3 model's capabilities.

Provenance

Source

Scripts created by uv-scripts for applying Meta's SAM3 model.

Collection Method

Scripts process external HuggingFace datasets using the SAM3 model.

Freshness

Last updated 2026-03-31.