SightRadar

List the face collections on this SightRadar account, with photo, face and selfie counts. Supports substring search and paging. Free.

Show one collection: status, photo_count, face_count, selfie_count, created_at and purge_after when a deletion is pending. Set verify=true to recount from the vector store instead of the cached counters. Free.

Create a face collection (for example one per event). Creating one that already exists and is active succeeds without change; one that is being deleted or was deleted returns 409 (use restore_collection or recreate_collection). Indexing into a missing collection also creates it, so this is optional. Free.

Delete a collection and every face, selfie and photo record in it. By default this is a SOFT delete: the collection is hidden at once and purged after a 7-day grace period, and restore_collection undoes it until then. immediate=true starts the purge now and cannot be undone. compliance=true records an audited erasure (GDPR/BIPA) and cannot be restored. Confirm with the user before calling this. Free.

Track a collection deletion: phase (not_deleting, deletion_pending, deleting, completed, blocked), whether the data is verified gone (verified_zero), purge_after, attempts and the last error. Free.

Undo a soft delete while the collection is still in its grace period (status deletion_pending). Fails once the purge has started or for a compliance erasure. Free.

Bring back an empty collection under the same id after it has been fully deleted (status deleted). The old faces are not recovered. Free.

Remove every face indexed from one photo. Soft by default (restore_photo undoes it); immediate=true or compliance=true is permanent. Deleting a photo that is already gone still succeeds. Free.

Undo a soft photo deletion made with delete_photo. Free.

Detect every face in a photo and store them in a collection so they can be found later by search_faces. Returns each face's point_id, bounding box (ratios of image width/height), detection score and pixel size. Faces below the quality gate (det_score < 0.6 or under 24 px) are counted as rejected. Indexing the same photo_id twice fails with already_indexed unless override=true. The collection is created if missing. Costs 93 credits.

Find the photos in a collection that contain the person in a query image (typically a selfie) or a 512-number embedding. Returns matches [{photo_id, score}] best first, plus photo_ids. When nothing matches, `reason` says why: no_face (no face found in the query) or low_quality_selfie. Costs 93 credits.

Find a registered person's photos again without re-sending their selfie: pass the point_id returned by register_selfie for this collection. Cheaper than search_faces. Only selfie point ids (a UUID or an unsigned integer) work here; face point ids from index_photo return reason=point_not_found. Returns matches [{photo_id, score}]. Costs 62 credits.

Store a user's selfie in a collection so they can be matched against its photos. Returns face_found, the selfie's point_id (use with search_by_point_id) and its 512-number embedding. face_found=false with reason=no_face means no face was found; a face that fails the quality gate (det_score < 0.6 or under 24 px) is refused with HTTP 422 and not stored. Retrying with the same idempotency_key returns the original point_id without a second charge. Costs 93 credits.

Find the faces in an image without storing anything. Returns detected_face_count, gated_face_count (faces that pass the quality gate) and per face: bounding box (ratios), det_score, min_px and quality_passed. Useful to check a photo or selfie before indexing. Costs 93 credits.

Check whether two images (or embeddings) show the same person. Uses the best face on each side. Returns face_found, similarity (0-1, null when a side has no face), match and the threshold used. Each side is a public URL, a CDN key or a 512-number embedding. Costs 93 credits per image processed: 186 for two images, 93 for one image and one embedding, 93 for two embeddings.

Queue up to 1000 photo URLs for background processing. op=index stores their faces in the collection; op=match matches them against the collection. Submitting is free; each photo costs 62 credits when it is processed (some accounts pay the real-time 93 for a batch of one photo). The collection is created if missing; one that is being deleted is refused with 409 collection_not_writable. Only URLs are accepted, not base64. Returns batch_id; poll get_batch or register a webhook to hear when it finishes. URLs are not checked at submit time: unreachable ones fail individually later.

List recent batch jobs newest first with status and succeeded/failed/pending counts. For the next page pass next_before_created and next_before_id from the previous answer. Free.

Status of one or many batch jobs: status, total_photos, succeeded, failed, pending. Pass batch_id for one, or batch_ids (up to 5000) for many at once; ids you don't own are left out. Free.

Per-photo results of a batch: external_id, status, face_count and reason_code for failures. Page with after_index = next_after_index. Free.

Retry batch photos that failed for a given reason, in two steps. First call with execute=false to get a preview: how many photos match, the estimated credit cost and a preview_id (valid 10 minutes, usable once). Show that to the user, then call again with execute=true and the preview_id. Requeued photos are charged when processed, usually 62 credits each; the preview's credits_per_photo and estimated_credits give the exact price.

List webhook endpoints (id, url, status, created_at). Optionally filter by status. Free.

Register a public HTTPS URL to receive batch-completion and deletion events. Each delivery is signed: X-SightRadar-Signature = hex HMAC-SHA256(secret, '<X-SightRadar-Timestamp>.<raw body>'). If you omit secret one is generated and returned ONCE in this answer: tell the user to store it. Free.

Stop deliveries to a webhook endpoint. Already-disabled endpoints succeed again. Free.

Deliveries that gave up after all retries (9 attempts): delivery_id, endpoint, url, attempts, last status code and error. Free.

Send failed deliveries again. Pass delivery_ids to replay specific ones, or omit to replay all of them. Returns how many were requeued. Free.

Current credit balance of the account. 100,000 credits = US$1. Real-time index, search, selfie, detect = 93 credits; compare = 93 per image; batch photo and search_by_point_id = 62. Free.

Takes no parameters.

Calls and credits spent, per operation. granularity=summary (default) gives totals per op; daily gives a day-by-day series including storage charges; storage gives the daily stored-vector count and storage credits. Use days (1-365) or a from/to date range (YYYY-MM-DD). Free.