3D Hair Reconstruction

Reconstructs 3D hair from a photo of a person.

1. Getting a token

Getting a token

2. Choice of photo

Choose a photo of a person.

Note: The quality of the photo must be a good resolution.

Note: Hair should be clearly visible.

Note: So far, we are unable to work well with ponytails and pigtails.

Need a successful image? Download a sample

3. Reconstruct 3D Hair

Make a request by sending your chosen photo. Make sure you are using the mode you really need. 'card' mode is used by default.

curl

token="ENTER_YOUR_TOKEN"
image_path="ENTER_IMAGE_PATH"

curl -X 'POST' \
  'https://api.metahumansdk.io/hair_recon/run_pipeline' \
  -H 'accept: application/json' \
  -H 'Content-Type: multipart/form-data' \
  -F "token=$token" \
  -F "image_bytes=@$image_path;type=image/png" > hair.glb \
  -F 'mode=card'

python

import requests

token = "ENTER_YOUR_TOKEN"
image_path = "ENTER_IMAGE_PATH"

if __name__ == "__main__":
    service_url = "https://api.metahumansdk.io/hair_recon"
    service_headers = {"accept": "application/json"}
    files = {
        "token": (None, token),
        "image_bytes": ("image", open(image_path, "rb"), "image/png"),
        "mode": (None, "card")  # card or volume
    }
    response_config = requests.post(service_url + "/run_pipeline", headers=service_headers, files=files, stream=True)
    with open("hair.glb", "wb") as f:
        f.write(response_config.content)

The expected result is the hair.glb file.

4. Check GLB model

Check saved file in GLB viewer (for example in Blender).

Examples of results

Open in Blender

Original photo

Reconstructed hair mesh in card mode

Reconstructed hair mesh in volume mode

Example rendering

Rendering in Blender

Rendered hair mesh in card mode

Rendered hair mesh in volume mode

5. Visualization with head

In order to get a 3D model of the head with hair, we will use GLB Constructor:

GLB Constructor currently supports only hair involume mode

At this step it is assumed that you have already received uv.png and head_config.jsonfrom 3D Face Reconstruction tab.

curl

token="ENTER_YOUR_TOKEN"

curl -X 'POST' 'https://api.metahumansdk.io/glb_const/assemble' \
    -H 'accept: application/json' \
    -H 'Content-Type: multipart/form-data' \
    -F "token=$token" \
    -F '[email protected];type=image/png' \
    -F 'head_config=@head_config.json;type=application/json' \
    -F '[email protected];type=image/png' \
    -F 'add_facs=true' > head.glb

python

import io
import json
import requests

token = "ENTER_YOUR_TOKEN"

if __name__ == "__main__":
    service_url = "https://api.metahumansdk.io/glb_const"
    service_headers = {'accept': 'application/json'}
    
    with open('uv.png', 'rb') as out:
        head_uv = out.read()
    with open('head_config.json', 'r') as out:
        head_config = json.loads(out.read())
    with open('hair.glb', 'rb') as out:
        custom_hair = out.read()

    files = {
        'token': (None, token),
        'head_uv': ('head_model', head_uv, 'image/png'),
        'head_config': ('head_config', json.dumps(head_config), 'application/json'),
        'custom_models': ('custom_models', io.BytesIO(custom_hair), 'application/json'),
        'add_facs': (None, True),
    }

    response = requests.post(service_url + "/assemble", headers=service_headers, files=files)
    assert (response.status_code == 200)
 
    with open("head.glb", "wb") as f:
        f.write(response.content)

After the completed stages, you get a glb model head.glb .

Result:

image.png

head.glb