> For the complete documentation index, see [llms.txt](https://docs.roboflow.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.roboflow.com/roboflow/roboflow-hi/deploy/supported-models/sam3.md). # SAM3 हम Meta के [Segment Anything Model 3](https://github.com/facebookresearch/sam3) के माध्यम से inferencing का समर्थन करते हैं [Serverless Hosted API](/roboflow/roboflow-hi/deploy/serverless-hosted-api-v2.md)। हम दो अलग-अलग SAM3 endpoints प्रदान करते हैं: {% hint style="info" %} Roboflow पर SAM3 model को train करना paid [plans](/roboflow/roboflow-hi/billing/plans.md) में उपलब्ध है, जिनमें [usage-based billing](/roboflow/roboflow-hi/billing/credits.md)शामिल है। इसके बाद, feature का उपयोग करने के लिए आप SAM3 architecture पर "Request Feature" button से access request कर सकते हैं [training flow](/roboflow/roboflow-hi/train/train.md). {% endhint %} * [Promptable concept segmentation](#concept-segmentation-pcs) (**PCS**), जो image में concept के हर instance को segment करता है। Concepts को text prompts, exemplar boxes, या दोनों के द्वारा वर्णित किया जाता है। * [Promptable visual segmentation](#visual-segmentation-pvs) (**PVS**), जो SAM2 की शैली में points या एक box से प्रति request एक object को interactive रूप से segment करता है। Endpoint चुनने के लिए इस table का उपयोग करें:
आपके पासआप चाहते हैंउपयोग करें
एक टेक्स्ट विवरण (उदा: "person")हर matching instance के लिए masks/sam3/concept_segment
एक उदाहरण object के चारों ओर एक boxहर similar instance के लिए masks/sam3/concept_segment
objects को include या exclude करने के लिए text plus example boxesहर matching instance के लिए masks/sam3/concept_segment
एक specific object पर एक click या एक boxकेवल उस object के लिए एक mask/sam3/visual_segment
अपनी [API key](https://app.roboflow.com/settings/api) को `api_key` query parameter के रूप में हर request में पास करें। ## Concept Segmentation (PCS) `POST https://serverless.roboflow.com/sam3/concept_segment` prompts में प्रत्येक entry `prompts` एक concept का वर्णन करती है। response में हर prompt के लिए एक `prompt_results` entry होती है, जिसमें मिले हुए सभी instances होते हैं। Requests अधिकतम 16 prompts स्वीकार करती हैं। ### Text prompts ```python import os import requests payload = { "image": {"type": "url", "value": "https://media.roboflow.com/inference/people-walking.jpg"}, "prompts": [ {"type": "text", "text": "person"}, {"type": "text", "text": "backpack"}, ], "output_prob_thresh": 0.5, "format": "polygon", # or "rle" } response = requests.post( "https://serverless.roboflow.com/sam3/concept_segment", params={"api_key": os.environ["ROBOFLOW_API_KEY"]}, json=payload, ) for prompt_result in response.json()["prompt_results"]: print(prompt_result["echo"], len(prompt_result["predictions"]), "instances") ``` Images को inline के रूप में भी भेजा जा सकता है `{"type": "base64", "value": ""}`. ### Exemplar box prompts Text के बजाय, आप एक exemplar के साथ prompt कर सकते हैं: एक उदाहरण object के चारों ओर एक box। Model सिर्फ boxed object नहीं, बल्कि example से मेल खाने वाले हर instance को ढूँढता है। ```python payload = { "image": {"type": "url", "value": "https://media.roboflow.com/inference/people-walking.jpg"}, "prompts": [ { "type": "visual", "boxes": [{"x": 1409, "y": 705, "width": 112, "height": 183}], "box_labels": [1], } ], "output_prob_thresh": 0.5, "format": "polygon", } ``` Boxes absolute pixel coordinates का उपयोग करते हैं। दो formats स्वीकार किए जाते हैं: * `{"x": ..., "y": ..., "width": ..., "height": ...}` जहाँ `x`, `y` ऊपरी-बाएँ कोना है * `{"x0": ..., "y0": ..., "x1": ..., "y1": ...}` स्पष्ट corners के लिए `box_labels` आवश्यक है जब `boxes` set किया गया हो और इसमें हर box के लिए एक entry होनी चाहिए: `1` एक positive exemplar को दर्शाता है (इस तरह के objects ढूँढें), `0` एक negative exemplar को दर्शाता है (इस तरह के objects को बाहर करें). ### संयुक्त text और exemplar prompts एक single prompt में text और exemplar boxes दोनों हो सकते हैं। यह visual examples के साथ किसी text concept को narrow करने, या negative exemplars के साथ lookalikes को exclude करने के लिए उपयोगी है: ```python payload = { "image": {"type": "url", "value": "https://media.roboflow.com/inference/people-walking.jpg"}, "prompts": [ { "type": "visual", "text": "person", "boxes": [ {"x": 1409, "y": 705, "width": 112, "height": 183}, {"x": 1216, "y": 496, "width": 124, "height": 184}, ], "box_labels": [1, 0], } ], "output_prob_thresh": 0.5, "format": "polygon", } ``` यहाँ model पहले (positive) exemplar से मेल खाने वाले लोगों को segment करता है, जबकि दूसरे (negative) exemplar से मिलते-जुलते instances को दबा देता है। ## Visual Segmentation (PVS) `POST https://serverless.roboflow.com/sam3/visual_segment` PVS clicks या एक box द्वारा इंगित एक specific object को segment करता है। इसका उपयोग interactive, human-in-the-loop mask refinement के लिए करें; जब आप किसी concept के हर instance को चाहते हों, तब PCS का उपयोग करें। ```python import os import requests payload = { "image": {"type": "url", "value": "https://media.roboflow.com/inference/people-walking.jpg"}, "prompts": { "prompts": [ { "points": [{"x": 1465, "y": 796, "positive": True}], "box": {"x": 1465, "y": 796, "width": 112, "height": 183}, } ] }, "multimask_output": False, "format": "json", } response = requests.post( "https://serverless.roboflow.com/sam3/visual_segment", params={"api_key": os.environ["ROBOFLOW_API_KEY"]}, json=payload, ) prediction = response.json()["predictions"][0] print(prediction["confidence"], len(prediction["masks"]), "polygons") ``` एक prompt में हो सकता है `points`का `box`या दोनों: * `points` absolute pixel coordinates होते हैं। `"positive": true` clicked region को शामिल करता है, `false` इसे बाहर करता है। mask को refine करने के लिए और points जोड़ें। * `box` center-anchored coordinates का उपयोग करता है: `x`, `y` box का center है, PCS boxes के विपरीत जो top-left anchored होते हैं। response में prompt के लिए single highest-confidence mask होता है। `multimask_output` यह नियंत्रित करता है कि model कितने internal mask proposals generate करता है (true होने पर तीन), लेकिन response के लिए हमेशा सबसे अच्छा proposal चुना जाता है। {% hint style="warning" %} हर request में एक prompt भेजें। एक PVS request में multiple prompts वर्तमान में केवल एक prediction लौटाते हैं। {% endhint %} OpenCV का उपयोग करके एक interactive demo के लिए, यह देखें [GitHub Gist](https://gist.github.com/Erol444/4cbc33c6ac52d83c63f6f9d86ca8a7a4)जिसका उपयोग इस video में किया गया था: {% embed url="" %} ## Inference speed Latency measured with [Roboflow Inference](https://inference.roboflow.com/) 1x NVIDIA L4 पर, batch size 1 के साथ, warmup के बाद mean।
ModelLatency (ms)
sam3251.4
एक single text prompt से concept segmentation के साथ मापा गया। ## Endpoints ## SAM3 PCS (promptable concept segmentation) > \*\*Concept Segmentation (Text Prompts)\*\*\ > \ > Allows you to segment objects using text prompts.\ > \ > \*\*Image Input\*\*: The \`image\` field accepts either:\ > \- \`{"type": "url", "value": "\"}\` - A publicly accessible image URL\ > \- \`{"type": "base64", "value": "\"}\` - Base64 encoded image data\ > \ > \*\*Prompts\*\*: Each prompt in the \`prompts\` array should have \`type: "text"\` and a \`text\` field with the object description. ```json {"openapi":"3.1.0","info":{"title":"Roboflow SAM3 API","version":"0.64.4"},"servers":[{"url":"https://serverless.roboflow.com"}],"paths":{"/sam3/concept_segment":{"post":{"summary":"SAM3 PCS (promptable concept segmentation)","description":"**Concept Segmentation (Text Prompts)**\n\nAllows you to segment objects using text prompts.\n\n**Image Input**: The `image` field accepts either:\n- `{\"type\": \"url\", \"value\": \"\"}` - A publicly accessible image URL\n- `{\"type\": \"base64\", \"value\": \"\"}` - Base64 encoded image data\n\n **Prompts**: Each prompt in the `prompts` array should have `type: \"text\"` and a `text` field with the object description.","operationId":"sam3_segment_image_sam3_concept_segment_post","parameters":[{"name":"api_key","in":"query","required":true,"schema":{"type":"string","title":"API Key"},"description":"Your Roboflow API Key. Get one at https://app.roboflow.com/settings/api"}],"requestBody":{"required":true,"content":{"application/json":{"schema":{"$ref":"#/components/schemas/Sam3SegmentationRequest"}}}},"responses":{"200":{"description":"Successful Response","content":{"application/json":{"schema":{"$ref":"#/components/schemas/Sam3SegmentationResponse"}}}},"422":{"description":"Validation Error","content":{"application/json":{"schema":{"$ref":"#/components/schemas/HTTPValidationError"}}}}}}}},"components":{"schemas":{"Sam3SegmentationRequest":{"properties":{"image":{"$ref":"#/components/schemas/InferenceRequestImage","description":"The image to be segmented."},"prompts":{"items":{"$ref":"#/components/schemas/Sam3Prompt"},"type":"array","minItems":1,"title":"Prompts","description":"List of prompts (text and/or visual)"},"format":{"type":"string","title":"Format","description":"One of 'polygon', 'rle'","default":"polygon"},"image_id":{"type":"string","title":"Image Id","description":"Optional ID for caching embeddings."},"output_prob_thresh":{"type":"number","title":"Output Prob Thresh","description":"Score threshold for outputs.","default":0.5},"model_id":{"type":"string","title":"Model Id","description":"The model ID of SAM3. Use 'sam3/sam3_final' to target the generic base model.","default":"sam3/sam3_final"},"nms_iou_threshold":{"type":"number","title":"Nms Iou Threshold","description":"IoU threshold for cross-prompt NMS. If not set, NMS is disabled. Must be in [0.0, 1.0] when set."}},"type":"object","required":["image","prompts"],"title":"Sam3SegmentationRequest"},"InferenceRequestImage":{"properties":{"type":{"type":"string","title":"Type","description":"The type of image data provided, one of `url`, `base64`"},"value":{"type":"string","title":"Value","description":"Image data corresponding to the image type, if type = 'url' then value is a string containing the url of an image, else if type = 'base64' then value is a string containing base64 encoded image data."}},"type":"object","required":["type"],"title":"InferenceRequestImage","description":"Image data for inference request.\n\nAttributes:\n type (str): The type of image data provided, one of 'url', 'base64', or 'numpy'.\n value (Optional[Any]): Image data corresponding to the image type."},"Sam3Prompt":{"properties":{"type":{"type":"string","title":"Type","description":"Hint: `text` or `visual`"},"text":{"type":"string","title":"Text","description":"Text prompt describing the object to segment"},"output_prob_thresh":{"type":"number","title":"Output Prob Thresh","description":"Score threshold for this prompt's outputs. Overrides request-level threshold if set."},"boxes":{"items":{"anyOf":[{"$ref":"#/components/schemas/Box"},{"$ref":"#/components/schemas/BoxXYXY"}]},"type":"array","title":"Boxes","description":"Absolute pixel boxes as either XYWH or XYXY entries"},"box_labels":{"items":{"anyOf":[{"type":"integer"},{"type":"boolean"}]},"type":"array","title":"Box Labels","description":"List of 0/1 or booleans for boxes"}},"type":"object","required":["type"],"title":"Sam3Prompt","description":"Unified prompt that can contain text and/or geometry. Absolute pixel coordinates are used for boxes."},"Sam3SegmentationResponse":{"properties":{"prompt_results":{"items":{"$ref":"#/components/schemas/Sam3PromptResult"},"type":"array","title":"Prompt Results","description":"Results for each prompt in the request"},"time":{"type":"number","title":"Time","description":"The time in seconds it took to produce the segmentation including preprocessing"}},"type":"object","required":["prompt_results","time"],"title":"Sam3SegmentationResponse"},"Sam3PromptResult":{"properties":{"prompt_index":{"type":"integer","title":"Prompt Index","description":"Index of the prompt this result corresponds to"},"echo":{"$ref":"#/components/schemas/Sam3PromptEcho","description":"Echo of the original prompt for reference"},"predictions":{"items":{"$ref":"#/components/schemas/Sam3SegmentationPrediction"},"type":"array","title":"Predictions","description":"Segmentation predictions for this prompt"}},"type":"object","required":["prompt_index","predictions"],"title":"Sam3PromptResult"},"Sam3PromptEcho":{"properties":{"prompt_index":{"type":"integer","title":"Prompt Index"},"type":{"type":"string","title":"Type","description":"The prompt type (`text` or `visual`)"},"text":{"type":"string","title":"Text","description":"The text prompt if type is `text`"},"num_boxes":{"type":"integer","title":"Num Boxes","description":"Number of bounding boxes in the prompt"}},"type":"object","title":"Sam3PromptEcho"},"Sam3SegmentationPrediction":{"properties":{"format":{"type":"string","title":"Format","description":"The format of the mask data, either `polygon` or `rle`"},"confidence":{"type":"number","title":"Confidence","description":"Confidence score for this prediction"},"masks":{"items":{"items":{"items":{"type":"number"},"type":"array","minItems":2,"maxItems":2},"type":"array"},"type":"array","title":"Masks","description":"Array of polygons, each polygon is an array of [x, y] coordinate points"}},"type":"object","required":["format","confidence","masks"],"title":"Sam3SegmentationPrediction"},"HTTPValidationError":{"properties":{"detail":{"items":{"$ref":"#/components/schemas/ValidationError"},"type":"array","title":"Detail"}},"type":"object","title":"HTTPValidationError"},"ValidationError":{"properties":{"loc":{"items":{"anyOf":[{"type":"string"},{"type":"integer"}]},"type":"array","title":"Location"},"msg":{"type":"string","title":"Message"},"type":{"type":"string","title":"Error Type"}},"type":"object","required":["loc","msg","type"],"title":"ValidationError"}}}} ``` ## SAM3 PVS (promptable visual segmentation) > \*\*Interactive Segmentation (SAM 2 Style)\*\*\ > \ > SAM 3 also supports interactive segmentation using points and boxes.\ > \ > \*\*Image Input\*\*: The \`image\` field accepts either:\ > \- \`{"type": "url", "value": "\"}\` - A publicly accessible image URL\ > \- \`{"type": "base64", "value": "\"}\` - Base64 encoded image data\ > \ > \> \*\*Note\*\*: NumPy arrays are NOT supported on the serverless API. Use URL or base64 encoding only.\ > \ > \*\*Prompts\*\*: Support point-based prompts with positive/negative clicks for interactive segmentation. ```json {"openapi":"3.1.0","info":{"title":"Roboflow SAM3 API","version":"0.64.4"},"servers":[{"url":"https://serverless.roboflow.com"}],"paths":{"/sam3/visual_segment":{"post":{"summary":"SAM3 PVS (promptable visual segmentation)","description":"**Interactive Segmentation (SAM 2 Style)**\n\nSAM 3 also supports interactive segmentation using points and boxes.\n\n**Image Input**: The `image` field accepts either:\n- `{\"type\": \"url\", \"value\": \"\"}` - A publicly accessible image URL\n- `{\"type\": \"base64\", \"value\": \"\"}` - Base64 encoded image data\n\n> **Note**: NumPy arrays are NOT supported on the serverless API. Use URL or base64 encoding only.\n\n**Prompts**: Support point-based prompts with positive/negative clicks for interactive segmentation.","operationId":"sam3_visual_segment_sam3_visual_segment_post","parameters":[{"name":"api_key","in":"query","required":true,"schema":{"type":"string","title":"API Key"},"description":"Your Roboflow API Key. Get one at https://app.roboflow.com/settings/api"}],"requestBody":{"required":true,"content":{"application/json":{"schema":{"$ref":"#/components/schemas/Sam2SegmentationRequest"}}}},"responses":{"200":{"description":"Successful Response","content":{"application/json":{"schema":{"$ref":"#/components/schemas/Sam2SegmentationResponse"}}}},"422":{"description":"Validation Error","content":{"application/json":{"schema":{"$ref":"#/components/schemas/HTTPValidationError"}}}}}}}},"components":{"schemas":{"Sam2SegmentationRequest":{"properties":{"image":{"$ref":"#/components/schemas/InferenceRequestImage","description":"The image to be segmented."},"image_id":{"type":"string","title":"Image Id","description":"The ID of the image to be segmented used to retrieve cached embeddings. If an embedding is cached, it will be used instead of generating a new embedding. If no embedding is cached, a new embedding will be generated and cached."},"prompts":{"$ref":"#/components/schemas/Sam2PromptSet","description":"A list of prompts for masks to predict. Each prompt can include a bounding box and / or a set of postive or negative points."},"format":{"type":"string","title":"Format","description":"The format of the response. Must be one of 'json', 'rle', or 'binary'. If binary, masks are returned as binary numpy arrays. If json, masks are converted to polygons. If rle, masks are converted to RLE format.","default":"json"},"sam2_version_id":{"type":"string","title":"Sam2 Version Id","description":"The version ID of SAM to be used for this request. Must be one of hiera_tiny, hiera_small, hiera_large, hiera_b_plus","default":"hiera_large"},"multimask_output":{"type":"boolean","title":"Multimask Output","description":"If true, the model will return three masks. For ambiguous input prompts (such as a single click), this will often produce better masks than a single prediction.","default":true},"save_logits_to_cache":{"type":"boolean","title":"Save Logits To Cache","description":"If True, saves the low-resolution logits to the cache for potential future use.","default":false},"load_logits_from_cache":{"type":"boolean","title":"Load Logits From Cache","description":"If True, attempts to load previously cached low-resolution logits for the given image and prompt set.","default":false}},"type":"object","required":["image"],"title":"Sam2SegmentationRequest","description":"SAM2 visual segmentation request."},"InferenceRequestImage":{"properties":{"type":{"type":"string","title":"Type","description":"The type of image data provided, one of `url`, `base64`"},"value":{"type":"string","title":"Value","description":"Image data corresponding to the image type, if type = 'url' then value is a string containing the url of an image, else if type = 'base64' then value is a string containing base64 encoded image data."}},"type":"object","required":["type"],"title":"InferenceRequestImage","description":"Image data for inference request.\n\nAttributes:\n type (str): The type of image data provided, one of 'url', 'base64', or 'numpy'.\n value (Optional[Any]): Image data corresponding to the image type."},"Sam2SegmentationResponse":{"properties":{"prompt_results":{"items":{"$ref":"#/components/schemas/Sam2PromptResult"},"type":"array","title":"Prompt Results","description":"Results for each prompt in the request"},"time":{"type":"number","title":"Time","description":"The time in seconds it took to produce the segmentation including preprocessing"}},"type":"object","required":["prompt_results","time"],"title":"Sam2SegmentationResponse"},"Sam2PromptResult":{"properties":{"prompt_index":{"type":"integer","title":"Prompt Index","description":"Index of the prompt this result corresponds to"},"predictions":{"items":{"$ref":"#/components/schemas/Sam2SegmentationPrediction"},"type":"array","title":"Predictions","description":"Segmentation predictions for this prompt"}},"type":"object","required":["prompt_index","predictions"],"title":"Sam2PromptResult"},"HTTPValidationError":{"properties":{"detail":{"items":{"$ref":"#/components/schemas/ValidationError"},"type":"array","title":"Detail"}},"type":"object","title":"HTTPValidationError"},"ValidationError":{"properties":{"loc":{"items":{"anyOf":[{"type":"string"},{"type":"integer"}]},"type":"array","title":"Location"},"msg":{"type":"string","title":"Message"},"type":{"type":"string","title":"Error Type"}},"type":"object","required":["loc","msg","type"],"title":"ValidationError"}}}} ```