Spaces:
Running
Running
| import numpy as np | |
| import mediapipe as mp | |
| from PIL import Image | |
| import PIL | |
| import scipy | |
| import scipy.ndimage | |
| mp_face_detection = mp.solutions.face_detection | |
| face_detection = mp_face_detection.FaceDetection(min_detection_confidence=0.5, model_selection=0) | |
| def aligns(pil_image,enable_padding=True,output_size=512,model_type="dlib",max_people=7): | |
| w,h = pil_image.size | |
| scale = 1 | |
| if min(w,h) > output_size*2: | |
| scale = min(w,h) / (output_size*2) | |
| new_w = int(w/scale) | |
| new_h = int(h/scale) | |
| pil_image = pil_image.resize((new_w,new_h),PIL.Image.BILINEAR) | |
| numpy_im = np.array(pil_image) | |
| #Find the locations of faces | |
| locations,context = get_locations(numpy_im,model_type)#face_recognition.face_locations(numpy_im) | |
| n_found = len(locations) | |
| print("Faces found",n_found) | |
| if (n_found == 0): | |
| return [] | |
| #How many are we going to return? | |
| n_to_return = min(n_found,max_people) | |
| #Return the largest ones | |
| areas = [(l[2] - l[0])*(l[1] - l[3]) for l in locations] | |
| indices = np.argpartition(areas, -n_to_return)[-n_to_return:] | |
| #Find the landmarks | |
| face_landmarks_list = get_landmarks(numpy_im,[locations[i] for i in indices],context,model_type)#face_recognition.face_landmarks(numpy_im,[locations[i]]) | |
| #Package them up | |
| to_return = [] | |
| for face in face_landmarks_list: | |
| im,quad = image_align(pil_image,face,enable_padding=enable_padding,output_size=output_size,transform_size=output_size) | |
| to_return.append((im,quad*scale)) | |
| #Return them | |
| return to_return | |
| def get_landmarks(numpy_array,locations,context,model_type="dlib"): | |
| ''' | |
| model_type can be "dlib" or "mediapipe" | |
| context is the second result from get_locations | |
| ''' | |
| assert(model_type in ["dlib","mediapipe"]) | |
| if model_type == "dlib": | |
| return face_recognition.face_landmarks(numpy_array,locations) | |
| else: | |
| return [context[tuple(l)] for l in locations] | |
| def landmarks_from_result(result,np_array): | |
| keypoint_names = ["left_eye", "right_eye", "nose" ,"mouth", "left_ear", "right_ear"] | |
| landmarks = {} | |
| for i,k in enumerate(result.location_data.relative_keypoints): | |
| x = round(k.x * np_array.shape[1]) | |
| y = round(k.y * np_array.shape[0]) | |
| landmarks[keypoint_names[i]] = np.array([x,y]) | |
| return landmarks | |
| def get_locations(numpy_array,model_type="dlib"): | |
| ''' | |
| model_type can be "dlib" or "mediapipe" | |
| returns face locations and a context for fast landmark finding | |
| ''' | |
| assert(model_type in ["dlib","mediapipe"]) | |
| if model_type == "dlib": | |
| return face_recognition.face_locations(numpy_array),None | |
| else: | |
| results = face_detection.process(np.array(numpy_array)) | |
| to_return = None | |
| im_h,im_w = numpy_array.shape[:2] | |
| box_list = [] | |
| landmarks = {} | |
| if results.detections is None: | |
| return box_list,landmarks | |
| for result in results.detections: | |
| x = round(result.location_data.relative_bounding_box.xmin*im_w) | |
| y = round(result.location_data.relative_bounding_box.ymin*im_h) | |
| w = round(result.location_data.relative_bounding_box.width*im_w) | |
| h = round(result.location_data.relative_bounding_box.height*im_h) | |
| box_list.append([x,y,x+w-1,y+h-1]) | |
| landmarks[(x,y,x+w-1,y+h-1)] = landmarks_from_result(result,numpy_array) | |
| return box_list,landmarks | |
| def align(pil_image,enable_padding=True,output_size=512,model_type="dlib"): | |
| w,h = pil_image.size | |
| scale = 1 | |
| if min(w,h) > output_size*2: | |
| scale = min(w,h) / (output_size*2) | |
| new_w = int(w/scale) | |
| new_h = int(h/scale) | |
| pil_image = pil_image.resize((new_w,new_h),PIL.Image.BILINEAR) | |
| numpy_im = np.array(pil_image) | |
| locations,context = get_locations(numpy_im,model_type)#face_recognition.face_locations(numpy_im) | |
| if (len(locations) == 0): | |
| return None | |
| areas = [(l[2] - l[0])*(l[1] - l[3]) for l in locations] | |
| i = np.argmax(areas) | |
| face_landmarks_list = get_landmarks(numpy_im,[locations[i]],context,model_type)#face_recognition.face_landmarks(numpy_im,[locations[i]]) | |
| im,quad = image_align(Image.fromarray(numpy_im),face_landmarks_list[0],enable_padding=enable_padding,output_size=output_size,transform_size=4*output_size) | |
| return im,quad*scale | |
| def image_align(img, lm, output_size=1024, transform_size=4096, enable_padding=True, x_scale=1, y_scale=1, em_scale=0.1, alpha=False): | |
| # Align function from FFHQ dataset pre-processing step | |
| # https://github.com/NVlabs/ffhq-dataset/blob/master/download_ffhq.py | |
| # Compute the land marks differently depending on what face finding model has been used | |
| if type(lm["left_eye"]) == np.ndarray and lm["left_eye"].size == 2: | |
| #Media pipe | |
| eye_left = lm["left_eye"] | |
| eye_right = lm["right_eye"] | |
| mouth_avg = lm["mouth"] | |
| else: | |
| #DLIB | |
| eye_left = np.mean(lm["left_eye"], axis=0) | |
| eye_right = np.mean(lm["right_eye"], axis=0) | |
| mouth_avg = (np.mean( lm["top_lip"],axis=0) + np.mean(lm["bottom_lip"],axis=0)) * 0.5 | |
| # Calculate auxiliary vectors. | |
| eye_avg = (eye_left + eye_right) * 0.5 | |
| eye_to_eye = eye_right - eye_left | |
| eye_to_mouth = mouth_avg - eye_avg | |
| # Choose oriented crop rectangle. | |
| x = eye_to_eye - np.flipud(eye_to_mouth) * [-1, 1] | |
| 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 means box | |
| qsize = np.hypot(*x) * 2 | |
| original_quad = np.copy(quad) | |
| # Load in-the-wild image. | |
| #img = img.convert('RGBA').convert('RGB') #I've already taken care of this | |
| # 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, PIL.Image.ANTIALIAS) | |
| 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 += (scipy.ndimage.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 = PIL.Image.fromarray(img, 'RGBA') | |
| else: | |
| img = PIL.Image.fromarray(img, 'RGB') | |
| quad += pad[:2] | |
| # Transform. | |
| img = img.transform((transform_size, transform_size), PIL.Image.QUAD, (quad + 0.5).flatten(), PIL.Image.BILINEAR) | |
| if output_size < transform_size: | |
| img = img.resize((output_size, output_size), PIL.Image.LANCZOS) | |
| # Save aligned image. | |
| return img,original_quad |