File size: 7,263 Bytes
40cfce6 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 |
# Copyright 2024-2025 The Alibaba Wan Team Authors. All rights reserved.
import os
import cv2
import math
import random
import numpy as np
def get_mask_boxes(mask):
"""
Args:
mask: [h, w]
Returns:
"""
y_coords, x_coords = np.nonzero(mask)
x_min = x_coords.min()
x_max = x_coords.max()
y_min = y_coords.min()
y_max = y_coords.max()
bbox = np.array([x_min, y_min, x_max, y_max]).astype(np.int32)
return bbox
def get_aug_mask(body_mask, w_len=10, h_len=20):
body_bbox = get_mask_boxes(body_mask)
bbox_wh = body_bbox[2:4] - body_bbox[0:2]
w_slice = np.int32(bbox_wh[0] / w_len)
h_slice = np.int32(bbox_wh[1] / h_len)
for each_w in range(body_bbox[0], body_bbox[2], w_slice):
w_start = min(each_w, body_bbox[2])
w_end = min((each_w + w_slice), body_bbox[2])
# print(w_start, w_end)
for each_h in range(body_bbox[1], body_bbox[3], h_slice):
h_start = min(each_h, body_bbox[3])
h_end = min((each_h + h_slice), body_bbox[3])
if body_mask[h_start:h_end, w_start:w_end].sum() > 0:
body_mask[h_start:h_end, w_start:w_end] = 1
return body_mask
def get_mask_body_img(img_copy, hand_mask, k=7, iterations=1):
kernel = np.ones((k, k), np.uint8)
dilation = cv2.dilate(hand_mask, kernel, iterations=iterations)
mask_hand_img = img_copy * (1 - dilation[:, :, None])
return mask_hand_img, dilation
def get_face_bboxes(kp2ds, scale, image_shape, ratio_aug):
h, w = image_shape
kp2ds_face = kp2ds.copy()[23:91, :2]
min_x, min_y = np.min(kp2ds_face, axis=0)
max_x, max_y = np.max(kp2ds_face, axis=0)
initial_width = max_x - min_x
initial_height = max_y - min_y
initial_area = initial_width * initial_height
expanded_area = initial_area * scale
new_width = np.sqrt(expanded_area * (initial_width / initial_height))
new_height = np.sqrt(expanded_area * (initial_height / initial_width))
delta_width = (new_width - initial_width) / 2
delta_height = (new_height - initial_height) / 4
if ratio_aug:
if random.random() > 0.5:
delta_width += random.uniform(0, initial_width // 10)
else:
delta_height += random.uniform(0, initial_height // 10)
expanded_min_x = max(min_x - delta_width, 0)
expanded_max_x = min(max_x + delta_width, w)
expanded_min_y = max(min_y - 3 * delta_height, 0)
expanded_max_y = min(max_y + delta_height, h)
return [int(expanded_min_x), int(expanded_max_x), int(expanded_min_y), int(expanded_max_y)]
def calculate_new_size(orig_w, orig_h, target_area, divisor=64):
target_ratio = orig_w / orig_h
def check_valid(w, h):
if w <= 0 or h <= 0:
return False
return (w * h <= target_area and
w % divisor == 0 and
h % divisor == 0)
def get_ratio_diff(w, h):
return abs(w / h - target_ratio)
def round_to_64(value, round_up=False, divisor=64):
if round_up:
return divisor * ((value + (divisor - 1)) // divisor)
return divisor * (value // divisor)
possible_sizes = []
max_area_h = int(np.sqrt(target_area / target_ratio))
max_area_w = int(max_area_h * target_ratio)
max_h = round_to_64(max_area_h, round_up=True, divisor=divisor)
max_w = round_to_64(max_area_w, round_up=True, divisor=divisor)
for h in range(divisor, max_h + divisor, divisor):
ideal_w = h * target_ratio
w_down = round_to_64(ideal_w)
w_up = round_to_64(ideal_w, round_up=True)
for w in [w_down, w_up]:
if check_valid(w, h, divisor):
possible_sizes.append((w, h, get_ratio_diff(w, h)))
if not possible_sizes:
raise ValueError("Can not find suitable size")
possible_sizes.sort(key=lambda x: (-x[0] * x[1], x[2]))
best_w, best_h, _ = possible_sizes[0]
return int(best_w), int(best_h)
def resize_by_area(image, target_area, keep_aspect_ratio=True, divisor=64, padding_color=(0, 0, 0)):
h, w = image.shape[:2]
try:
new_w, new_h = calculate_new_size(w, h, target_area, divisor)
except:
aspect_ratio = w / h
if keep_aspect_ratio:
new_h = math.sqrt(target_area / aspect_ratio)
new_w = target_area / new_h
else:
new_w = new_h = math.sqrt(target_area)
new_w, new_h = int((new_w // divisor) * divisor), int((new_h // divisor) * divisor)
interpolation = cv2.INTER_AREA if (new_w * new_h < w * h) else cv2.INTER_LINEAR
resized_image = padding_resize(image, height=new_h, width=new_w, padding_color=padding_color,
interpolation=interpolation)
return resized_image
def padding_resize(img_ori, height=512, width=512, padding_color=(0, 0, 0), interpolation=cv2.INTER_LINEAR):
ori_height = img_ori.shape[0]
ori_width = img_ori.shape[1]
channel = img_ori.shape[2]
img_pad = np.zeros((height, width, channel))
if channel == 1:
img_pad[:, :, 0] = padding_color[0]
else:
img_pad[:, :, 0] = padding_color[0]
img_pad[:, :, 1] = padding_color[1]
img_pad[:, :, 2] = padding_color[2]
if (ori_height / ori_width) > (height / width):
new_width = int(height / ori_height * ori_width)
img = cv2.resize(img_ori, (new_width, height), interpolation=interpolation)
padding = int((width - new_width) / 2)
if len(img.shape) == 2:
img = img[:, :, np.newaxis]
img_pad[:, padding: padding + new_width, :] = img
else:
new_height = int(width / ori_width * ori_height)
img = cv2.resize(img_ori, (width, new_height), interpolation=interpolation)
padding = int((height - new_height) / 2)
if len(img.shape) == 2:
img = img[:, :, np.newaxis]
img_pad[padding: padding + new_height, :, :] = img
img_pad = np.uint8(img_pad)
return img_pad
def get_frame_indices(frame_num, video_fps, clip_length, train_fps):
start_frame = 0
times = np.arange(0, clip_length) / train_fps
frame_indices = start_frame + np.round(times * video_fps).astype(int)
frame_indices = np.clip(frame_indices, 0, frame_num - 1)
return frame_indices.tolist()
def get_face_bboxes(kp2ds, scale, image_shape):
h, w = image_shape
kp2ds_face = kp2ds.copy()[1:] * (w, h)
min_x, min_y = np.min(kp2ds_face, axis=0)
max_x, max_y = np.max(kp2ds_face, axis=0)
initial_width = max_x - min_x
initial_height = max_y - min_y
initial_area = initial_width * initial_height
expanded_area = initial_area * scale
new_width = np.sqrt(expanded_area * (initial_width / initial_height))
new_height = np.sqrt(expanded_area * (initial_height / initial_width))
delta_width = (new_width - initial_width) / 2
delta_height = (new_height - initial_height) / 4
expanded_min_x = max(min_x - delta_width, 0)
expanded_max_x = min(max_x + delta_width, w)
expanded_min_y = max(min_y - 3 * delta_height, 0)
expanded_max_y = min(max_y + delta_height, h)
return [int(expanded_min_x), int(expanded_max_x), int(expanded_min_y), int(expanded_max_y)] |