Image

Location Extractor

High-precision place recognition, landmark detection, and geographic inference for images and video

0K runs

Note: This playground provides simulated output to showcase functionality. No input data is processed or stored on our servers. Use this demo to explore the feature extractor's capabilities before integrating it into your application.

Input

File URL string

Enter a URL to a image file

Upload image

Drag and drop a image file here, or click to browse

Select File

# video_url string

URL pointing to video file. Supported formats: MP4, MOV, AVI. Default: undefined

# image_url string

URL pointing to image file. Supported formats: JPG, PNG, WEBP. Default: undefined

# extract_every_n_frames integer

Video sampling rate (frames to skip between extractions). Default: 30

# min_confidence number

Minimum confidence threshold for location candidates (0.0-1.0). Default: 0.25

# candidate_limit integer

Maximum number of location candidates to return per frame. Default: 5

# mode string

Processing mode for accuracy vs performance. Default: standard

# enable_scene_embedding boolean

Toggle SigLIP2 semantic scene embeddings. Default: true

# enable_verification boolean

Enable LoFTR geometric verification for re-ranking. Default: false

# vector_dimension integer

Combined embedding dimension size. Default: 2048

# extract_indoor_outdoor boolean

Classify location as indoor or outdoor. Default: true

# extract_candidates boolean

Generate location name candidates. Default: true

# scene_embedding_weight number

Weight for semantic embeddings in fusion (0.0-1.0). Default: 0.3

# verification_top_k integer

Number of candidates to verify with LoFTR. Default: 5

# geometric_threshold number

Minimum LoFTR score for verification pass. Default: 0.65

Output

{
  "document_id": "doc_abc123",
  "collection_id": "col_xyz789",
  "source_object_id": "obj_def456",
  "location": {
    "place_descriptor": {
      "vector": [
        0.12,
        0.34,
        -0.21
      ],
      "model": "megaloc-v1",
      "dimension": 1024,
      "normalized": true
    },
    "scene_descriptor": {
      "vector": [
        0.08,
        -0.15,
        0.42
      ],
      "model": "siglip2-so400m",
      "dimension": 1024,
      "normalized": true
    },
    "fused_vector": [
      0.11,
      0.19,
      0.05
    ],
    "location_candidates": [
      {
        "name": "Eiffel Tower",
        "type": "landmark",
        "city": "Paris",
        "country": "France",
        "score": 0.87,
        "verified": true,
        "verification_score": 0.82
      },
      {
        "name": "Paris, France",
        "type": "city",
        "region": "Île-de-France",
        "country": "France",
        "score": 0.75,
        "verified": false
      }
    ],
    "environment": {
      "indoor_outdoor": "outdoor",
      "confidence": 0.94
    }
  },
  "frame": {
    "number": 120,
    "timestamp": 4,
    "width": 1920,
    "height": 1080
  },
  "video": {
    "filename": "travel_paris.mp4",
    "duration": 180.5,
    "fps": 30
  }
}

Ready to run Location Extractor on your data? Spin it up in Studio — no infra to host.

Run this in Studio

Already have embeddings? Skip extraction — search your own vectors with MVS. First 1M vectors free.

Try MVS →