face_detection.py

import insightface
import numpy as np
from PIL import Image, ImageOps
import cv2
from scipy.ndimage import gaussian_filter

# Load InsightFace models
try:
    detector = insightface.model_zoo.get_model('retinaface_mnet025_v2')
    detector.prepare(ctx_id=-1)  # Use CPU, set ctx_id=0 for GPU
except Exception as e:
    print(f"Error loading face detection model: {e}")
    detector = None

try:
    landmark_model = insightface.model_zoo.get_model('2d106det')
    landmark_model.prepare(ctx_id=-1)  # Use CPU, set ctx_id=0 for GPU
except Exception as e:
    print(f"Error loading landmark model: {e}")
    landmark_model = None

def align(image_in, face_index=0, output_size=256):
    if detector is None or landmark_model is None:
        print("Models not loaded properly.")
        return image_in, 0, None

    try:
        image_in = ImageOps.exif_transpose(image_in)
    except:
        print("Exif problem, not rotating")

    landmarks = list(get_landmarks(image_in))
    n_faces = len(landmarks)
    face_index = min(n_faces-1, face_index)
    if n_faces == 0:
        aligned_image = image_in
        quad = None
    else:
        aligned_image, quad = image_align(image_in, landmarks[face_index], output_size=output_size)

    return aligned_image, n_faces, quad

def get_landmarks(image):
    """Get landmarks from PIL image"""
    if detector is None or landmark_model is None:
        return []

    img = np.array(image)
    bboxes, _ = detector.detect(img, threshold=0.5, scale=1.0)

    for bbox in bboxes:
        # Use the landmark model to predict landmarks
        landmarks = landmark_model.get(img, bbox)
        yield landmarks

def image_align(src_img, face_landmarks, output_size=512, transform_size=2048, enable_padding=True, x_scale=1, y_scale=1, em_scale=0.1, alpha=False):
    # Align function modified from ffhq-dataset
    # See https://github.com/NVlabs/ffhq-dataset for license

    lm = np.array(face_landmarks)
    lm_eye_left = lm[2:3]  # left-clockwise
    lm_eye_right = lm[0:1]  # left-clockwise

    # Calculate auxiliary vectors.
    eye_left = np.mean(lm_eye_left, axis=0)
    eye_right = np.mean(lm_eye_right, axis=0)
    eye_avg = (eye_left + eye_right) * 0.5
    eye_to_eye = 0.71 * (eye_right - eye_left)
    mouth_avg = lm[4]
    eye_to_mouth = 1.35 * (mouth_avg - eye_avg)

    # Choose oriented crop rectangle.
    x = eye_to_eye.copy()
    x /= np.hypot(*x)
    x *= max(np.hypot(*eye_to_eye) * 2.0, np.hypot(*eye_to_mouth) * 1.8)
    x *= x_scale
    y = np.flipud(x) * [-y_scale, y_scale]
    c = eye_avg + eye_to_mouth * em_scale
    quad = np.stack([c - x - y, c - x + y, c + x + y, c + x - y])
    quad_orig = quad.copy()
    qsize = np.hypot(*x) * 2

    img = src_img.convert('RGBA').convert('RGB')

    # Shrink.
    shrink = int(np.floor(qsize / output_size * 0.5))
    if shrink > 1:
        rsize = (int(np.rint(float(img.size[0]) / shrink)), int(np.rint(float(img.size[1]) / shrink)))
        img = img.resize(rsize, Image.Resampling.LANCZOS)
        quad /= shrink
        qsize /= shrink

    # Crop.
    border = max(int(np.rint(qsize * 0.1)), 3)
    crop = (int(np.floor(min(quad[:, 0]))), int(np.floor(min(quad[:, 1]))), int(np.ceil(max(quad[:, 0]))), int(np.ceil(max(quad[:, 1]))))
    crop = (max(crop[0] - border, 0), max(crop[1] - border, 0), min(crop[2] + border, img.size[0]), min(crop[3] + border, img.size[1]))
    if crop[2] - crop[0] < img.size[0] or crop[3] - crop[1] < img.size[1]:
        img = img.crop(crop)
        quad -= crop[0:2]

    # Pad.
    pad = (int(np.floor(min(quad[:, 0]))), int(np.floor(min(quad[:, 1]))), int(np.ceil(max(quad[:, 0]))), int(np.ceil(max(quad[:, 1]))))
    pad = (max(-pad[0] + border, 0), max(-pad[1] + border, 0), max(pad[2] - img.size[0] + border, 0), max(pad[3] - img.size[1] + border, 0))
    if enable_padding and max(pad) > border - 4:
        pad = np.maximum(pad, int(np.rint(qsize * 0.3)))
        img = np.pad(np.float32(img), ((pad[1], pad[3]), (pad[0], pad[2]), (0, 0)), 'reflect')
        h, w, _ = img.shape
        y, x, _ = np.ogrid[:h, :w, :1]
        mask = np.maximum(1.0 - np.minimum(np.float32(x) / pad[0], np.float32(w - 1 - x) / pad[2]), 1.0 - np.minimum(np.float32(y) / pad[1], np.float32(h - 1 - y) / pad[3]))
        blur = qsize * 0.02
        img += (gaussian_filter(img, [blur, blur, 0]) - img) * np.clip(mask * 3.0 + 1.0, 0.0, 1.0)
        img += (np.median(img, axis=(0, 1)) - img) * np.clip(mask, 0.0, 1.0)
        img = np.uint8(np.clip(np.rint(img), 0, 255))
        if alpha:
            mask = 1 - np.clip(3.0 * mask, 0.0, 1.0)
            mask = np.uint8(np.clip(np.rint(mask * 255), 0, 255))
            img = np.concatenate((img, mask), axis=2)
            img = Image.fromarray(img, 'RGBA')
        else:
            img = Image.fromarray(img, 'RGB')
        quad += pad[:2]

    # Transform.
    img = img.transform((transform_size, transform_size), Image.QUAD, (quad + 0.5).flatten(), Image.BILINEAR)
    if output_size < transform_size:
        img = img.resize((output_size, output_size), Image.Resampling.LANCZOS)

    return img, quad_orig