Spatial Vision

Use non-LLM spatial VQA and grounded pointing with explicit degradation checks.

The canonical Spatial Vision endpoint and its compatibility alias are:

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

Both routes provide spatial yes/no answers and normalized object-relative pointing without invoking a language model. Use /vision/spatial for new integrations; /vision/query is the compatibility alias.

Spatial Vision requests are stateless. Send the question and image directly, with depth when the task uses it. You do not need a Domain, events, feedback, or learner configuration. When detector and segmenter processing is enabled, callers do not need to provide object hints or scene masks.

curl -fsS -X POST https://rei-neuroadapt-api.reilabs.org/api/v1/vision/spatial \
  -H "Authorization: Bearer <api-key>" \
  -H "Content-Type: application/json" \
  -d '{"question":"Pinpoint empty space to the left of the bowl.","image_base64":"<base64-image>","use_detector":true,"use_segmenter":true}'

Request contract

Send a required question and one image field:

• image, image_base64, or image_url: base64 bytes or a data URL;
• depth or depth_base64: optional depth input in the same representation;
• object_hints: optional caller-supplied normalized boxes;
• scene_masks: optional caller-supplied masks;
• use_detector and use_segmenter: grounding controls.

Despite its compatibility name, image_url must be a base64 value or data URL, not an arbitrary remote URL.

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

Validate grounded output

Returned points and bounding boxes are normalized to [0, 1]. Inspect answer, confidence, points, rationale, and the parsed query fields.

A successful HTTP response can still be a degraded fallback. For object-relative pointing, treat output as grounded only when:

vision.detector.status === "ok"
vision.detector.region_count > 0

When segmentation is needed, also inspect vision.segmenter.status and vision.segmenter.mask_count.