Segment Anything 2 Model

Convert bounding boxes to polygons, or run SAM2 on an entire image to generate a mask.

Run Segment Anything 2, a zero-shot instance segmentation model, on an image.

** Dedicated inference server required (GPU recomended) **

You can use pass in boxes/predictions from other models to Segment Anything 2 to use as prompts for the model. If you pass in box detections from another model, the class names of the boxes will be forwarded to the predicted masks. If using the model unprompted, the model will assign integers as class names / ids.

Type identifier

Use the following identifier in step "type" field: roboflow_core/segment_anything@v1 to add the block as a step in your workflow.

Properties

Name	Type	Description	Refs
name	str	Enter a unique identifier for this step..	❌
version	str	Model to be used. One of hiera_large, hiera_small, hiera_tiny, hiera_b_plus.	✅
threshold	float	Threshold for predicted masks scores.	✅
multimask_output	bool	Flag to determine whether to use sam2 internal multimask or single mask mode. For ambiguous prompts setting to True is recomended..	✅

The Refs column marks possibility to parametrise the property with dynamic values available in workflow runtime. See Bindings for more info.

Runtime compatibility

hard - runtime self_hosted_cpu; execution local : Requires a GPU; run_locally() loads a model that needs CUDA.

Input and Output Bindings

The available connections depend on its binding kinds. Check what binding kinds Segment Anything 2 Model in version v1 has.

Input and output bindings

• input

  • images (image): The image to infer on..

  • boxes (Union[instance_segmentation_prediction, keypoint_detection_prediction, object_detection_prediction]): Bounding boxes (from another model) to convert to polygons.

  • version (string): Model to be used. One of hiera_large, hiera_small, hiera_tiny, hiera_b_plus.

  • threshold (float): Threshold for predicted masks scores.

  • multimask_output (boolean): Flag to determine whether to use sam2 internal multimask or single mask mode. For ambiguous prompts setting to True is recomended..

• output

  • predictions (instance_segmentation_prediction): Prediction with detected bounding boxes and segmentation masks in form of sv.Detections(...) object.

Example JSON definition

{
	    "name": "<your_step_name_here>",
	    "type": "roboflow_core/segment_anything@v1",
	    "images": "$inputs.image",
	    "boxes": "$steps.object_detection_model.predictions",
	    "version": "hiera_large",
	    "threshold": 0.3,
	    "multimask_output": true
	}