Best Solution for Face Parsing (Segmentation)

[XavierJiezou]

Demo

• Input
  
• Output
  

Install

pip install torch pyfacer==0.04

Code

import torch
import facer
import numpy as np
from PIL import Image


def save_result(mask, output_path):
    palette = np.array(
        [
            [0, 0, 0],        # "background"
            [255, 153, 51],   # "neck"
            [204, 0, 0],      # "face"
            [0, 204, 0],      # "cloth"
            [102, 51, 0],     # "rr"
            [255, 0, 0],      # "lr"
            [0, 255, 255],    # "rb"
            [255, 204, 204],  # "lb"
            [51, 51, 255],    # "re"
            [204, 0, 204],    # "le"
            [76, 153, 0],     # "nose"
            [102, 204, 0],    # "imouth"
            [0, 0, 153],      # "llip"
            [255, 255, 0],    # "ulip"
            [0, 0, 204],      # "hair"
            [204, 204, 0],    # "eyeg"
            [255, 51, 153],   # "hat"
            [0, 204, 204],    # "earr"
            [0, 51, 0],       # "neck_l"
        ],
        dtype=np.uint8,
    )
    mask = mask.squeeze(0).cpu().byte().numpy()
    mask = Image.fromarray(mask, mode="P")
    mask.putpalette(palette.flatten())
    mask.save(output_path)


def inference(input_path, output_path):
    device = "cuda" if torch.cuda.is_available() else "cpu"

    image = facer.hwc2bchw(facer.read_hwc(input_path)).to(
        device
    )  # image: 1 x 3 x h x w

    model = facer.face_parser(
        "farl/celebm/448",
        device,
        model_path="https://github.com/FacePerceiver/facer/releases/download/models-v1/face_parsing.farl.celebm.main_ema_181500_jit.pt",
    )

    with torch.inference_mode():
        logits, _ = model.net(image / 255.0)
        mask = logits.argmax(dim=1)

    save_result(mask, output_path)


if __name__ == "__main__":
    inference("input.jpg", "output.png")

[yinglinzheng]

Thank your for the code snippet, this behavior now is better supported in pyfacer==0.0.5

import torch.nn.functional as F

# on already face aligned image, for example FFHQ dataset or CelebA aligned image
images = facer.hwc2bchw(facer.read_hwc('data/ffhq_15723.jpg')).to(device=device)  # image: 1 x 3 x h x w

facer.show_bchw(images)

# do some preprocessing
images = images.to(dtype=torch.float32) / 255.0 # [0, 1]
images = F.interpolate(images, size=(448, 448), mode='bilinear', align_corners=False) # as it's already face aligned, directly resize to 448x448

seg_logits, seg_preds, label_names = face_parser.forward_warped(images)

print(label_names)

plt.imshow(seg_preds[0].cpu().numpy())

checkout sample for more details.