To search by external query image, upload external the image and make it
available as a query image by issuing the following call: 
POST /api/v1/dataset/{dataset-id}/search-image-similarity?entity_type=IMAGES&bounding_box=X,Y,WIDTH,HEIGHT
Headers: Authorization: Bearer <jwt>

​
Request Parameters:

  • file: Query image file
  • bounding_box (optional): X, Y, Width, Height

​
Visual Layer API - Dataset Exploration

After uploading the image you can call exploration endpoint with anchor media id: GET /api/v1/explore/{dataset-id} 
GET /api/v1/explore/{dataset-id}?threshold=0&anchor_type={anchor_type}&entity_type={ENTITY_TYPE}&page_number={PAGE_NUMBER}&anchor_media_id={anchor_media_id}
Headers: Authorization: Bearer <jwt>

​
Query Parameters

  • entity_type IMAGES or OBJECTS
  • anchor_type MEDIA or UPLOAD for when you uploaded the image using the upload endpoint
  • anchor_media_id id of the image you want to use as anchor, returned from the upload for search endpoint
  • threshold similarity threshold(0-4) the search results are clustered according to this value
  • PAGE_NUMBER: Which page of results to view (e.g., 0 for start)

​
relevance_score

Each result item includes a relevance_score, which measures similarity to the anchor image.
  • Scale:
    0 = Most similar
    2 = Least similar
  • This score is based on cosine distance and is available per media item in the response.

​
Sample Response

  {
    "clusters": [
        {
            "cluster_id": "bffbc083-f5a6-426d-a9cf-c0b8a3ed232a",
            "type": "IMAGES",
            "n_images": 7,
            "n_objects": -1,
            "n_videos": -1,
            "n_frames": -1,
            "size_display_value": "0B",
            "similarity_threshold": "1",
            "previews": [
                {
                    "type": "IMAGE",
                    "media_id": "8c...e4c",
                    "media_uri": "/.../e4c.jpg",
                    "media_thumb_uri": "/.../e4c.jpg.webp",
                    "caption": "The image presents..",
                    "image_id": "8c...e4c",
                    "image_uri": "/..3e4c.jpg",
                    "bounding_box": null,
                    "file_name": "5049.tif",
                    "original_uri": "visual-layer/datasets/huji-geo/5049.tif",
                    "video_uri": "",
                    "frame_timestamp": null,
                    "labels": null,
                    "user_tags": null,
                    "relevance_score": 0.1304212444996986,
                    "relevance_score_type": "cosine_distance"
                }
      ],
    "metadata": {
        "used_duckdb": false
    }
}

​
Example:

curl -H "Authorization: Bearer <jwt>" https://app.visual-layer.com/api/v1/explore/{dataset-id}?threshold=0&entity_type=IMAGES&page_number=0&anchor_media_id={anchor_media_id}

​
Notes:

  • Results are paginated with a fixed page size of 100 clusters per page
  • Higher threshold values (0-4 depending on the dataset granularity options) create more granular clusters with stricter similarity requirements
  • When using anchor-based similarity search, results are ordered by similarity to the anchor
  • The entity_type parameter controls whether to return image clusters, object clusters, or both
  • For similarity search with an uploaded image, first use the /search-image-similarity endpoint
More detailed example is here