Spatial Vision endpoint

Run non-LLM spatial VQA and grounded object-relative pointing.

Endpoint

POST https://rei-neuroadapt-api.reilabs.org/api/v1/vision/spatial

The compatibility alias is POST https://rei-neuroadapt-api.reilabs.org/api/v1/vision/query. Use the canonical /vision/spatial route for new integrations.

Request

{
  "question": "Pinpoint empty space to the left of the bowl.",
  "image_base64": "<base64-image>",
  "use_detector": true,
  "use_segmenter": true
}

Accepted image fields are image, image_base64, and image_url; each contains base64 or a data URL, not an arbitrary remote URL. Optional depth fields are depth and depth_base64. You can also provide object_hints or scene_masks.

Validate the result

Points and bounding boxes are normalized to [0, 1]. A 200 response can still be a degraded fallback. For object-relative pointing, require vision.detector.status to be ok and vision.detector.region_count to be positive; inspect segmenter status and mask count when segmentation is required.

Related

See Spatial Vision for integration guidance and degradation handling.