Face Detection

Finds the faces in a photograph and returns each one's bounding box, a confidence score and the face itself as a Base64 crop.

POST/v3/store/ekyc/face-detection/single

+1 endpoints

Try it

• Price0.2 ICper request
• Measured85.6%
• Served565.8Kcalls

• v1.0
• Active
• Benchmark paper (PDF)

Input

OutputPOST /v3/store/ekyc/face-detection/single

bbox

xmax: 320.15 xmin: 100.23 ymax: 450.78 ymin: 150.45

detection_score

0.9998989105224609

face

/9j/4AAQ...base64_encoded_image

message

success

process_time

0.41

Every field, under ResponseOther endpoints: /multi see Request

Request

curl -X POST https://api.iapp.co.th/v3/store/ekyc/face-detection/single \
  -H "apikey: YOUR_API_KEY" \
  -F "file=@photo.jpg"

Response

200OK. The fields are explained under Response.

{
    "bbox": {
        "xmax": 320.15,
        "xmin": 100.23,
        "ymax": 450.78,
        "ymin": 150.45
    },
    "detection_score": 0.9998989105224609,
    "face": "/9j/4AAQ...base64_encoded_image",
    "message": "success",
    "process_time": 0.41
}

Request

multipart/form-data, one image per request.

Headers

apikeystringrequired

Your API key, created under API Keys in the control panel.

Body

filefilerequired

The photograph. JPEG or PNG, up to 10 MB.

Endpoint	Returns	Price
/v3/store/ekyc/face-detection/single	The most prominent face	0.2 IC per request
/v3/store/ekyc/face-detection/multi	Every face it detects, as result[]	0.3 IC per request

The legacy paths /face_detect_single and /face_detect_multi answer at the same prices.

Response

/single returns one face at the top level, as in the panel; /multi returns the same fields for each face inside result.

bboxobject

The face's box, in pixels of the input image

bbox.xminnumber

bbox.yminnumber

bbox.xmaxnumber

bbox.ymaxnumber

detection_scorenumber

Confidence, 0 to 1

facestring

The face crop, Base64 JPEG

messageprocess_time

success; seconds

Details

Measured

On the production service, 28 October 2026, through /multi, against the full validation set of the public WIDER FACE benchmark: 3,226 everyday photographs, from portraits to crowds, scored with the benchmark's official evaluation method.

Measure	Result
Faces at least 64 px tall found, of 5,109	85.6%
Faces 32 to 63 px tall found	66.3%
Faces 16 to 31 px tall found	21.2%
Returned boxes that matched a labelled face	98.9%
Average precision, Easy / Medium / Hard subset	80.5% / 72.6% / 37.0%
Processing time per image, median / 90th percentile	0.37 s / 1.12 s
Median time through the API	0.58 s

The service returns only faces it is highly confident about, so every returned box can be used without further filtering; a face below that confidence is not returned. For selfies, identity-card photos and portraits, where the face is large, the 64-pixel row applies; the Hard subset is dominated by faces under 32 pixels, which the service mostly misses. Method and results by face size: the benchmark (PDF).

Limits

• Faces under 32 pixels tall are mostly missed: 21.2% were found at 16 to 31 pixels and 0.7% below 16. The service is not designed for counting people in crowds; /multi returned as many as 101 faces for one test image.
• Only faces above a fixed confidence are returned; every box in the test had a detection_score of 0.95 or more. Low resolution, poor lighting or focus, head pose and occlusion can put a face below it.

Data handling

Images are processed in memory and not kept after the response. iApp keeps no copy; consent and notice under PDPA and GDPR remain with you as the data controller.

Changelog

Version	Date	Changes
1.0	Jan 2023	First release.