File size: 14,143 Bytes
3bbb319 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 |
# -*- coding: utf-8 -*-
# Max-Planck-Gesellschaft zur Förderung der Wissenschaften e.V. (MPG) is
# holder of all proprietary rights on this computer program.
# You can only use this computer program if you have closed
# a license agreement with MPG or you get the right to use the computer
# program from someone who is authorized to grant you that right.
# Any use of the computer program without a valid license is prohibited and
# liable to prosecution.
#
# Copyright©2019 Max-Planck-Gesellschaft zur Förderung
# der Wissenschaften e.V. (MPG). acting on behalf of its Max Planck Institute
# for Intelligent Systems. All rights reserved.
#
# Contact: ps-license@tuebingen.mpg.de
import os
import cv2
import numpy as np
import os.path as osp
from sklearn.random_projection import johnson_lindenstrauss_min_dim
import torch
from torch.utils.data import Dataset
from torchvision.transforms.functional import to_tensor
import torch.nn.functional as F
from torchvision.transforms import Normalize
from utils.smooth_bbox import get_all_bbox_params
from .data_utils.img_utils import get_single_image_crop_demo
from utils.cam_params import read_cam_params, homo_vector
from pymaf_core import path_config, constants
from pymaf_core.cfgs import cfg
from utils.imutils import crop, flip_img, flip_pose, flip_aa, flip_kp, transform, get_transform, get_rot_transf, rot_aa
class Inference(Dataset):
def __init__(self, image_folder, frames, bboxes=None, joints2d=None, scale=1.0, crop_size=224, pre_load_imgs=None, full_body=False, person_ids=[], wb_kps={}):
self.pre_load_imgs = pre_load_imgs
if pre_load_imgs is None:
self.image_file_names = [
osp.join(image_folder, x)
for x in os.listdir(image_folder)
if x.endswith('.png') or x.endswith('.jpg')
]
self.image_file_names = sorted(self.image_file_names)
self.image_file_names = np.array(self.image_file_names)[frames]
self.bboxes = bboxes
self.joints2d = joints2d
self.scale_factor = scale
self.crop_size = crop_size
self.frames = frames
self.has_keypoints = True if joints2d is not None else False
self.full_body = full_body
self.person_ids = person_ids
self.normalize_img = Normalize(mean=constants.IMG_NORM_MEAN, std=constants.IMG_NORM_STD)
self.norm_joints2d = np.zeros_like(self.joints2d)
if self.has_keypoints:
if not self.full_body:
bboxes, time_pt1, time_pt2 = get_all_bbox_params(joints2d, vis_thresh=0.3)
bboxes[:, 2:] = 150. / bboxes[:, 2:]
self.bboxes = np.stack([bboxes[:, 0], bboxes[:, 1], bboxes[:, 2], bboxes[:, 2]]).T
self.image_file_names = self.image_file_names[time_pt1:time_pt2]
self.joints2d = joints2d[time_pt1:time_pt2]
self.frames = frames[time_pt1:time_pt2]
else:
bboxes = []
scales = []
for j2d in joints2d:
kp2d_valid = j2d[j2d[:, 2]>0.]
bbox = [min(kp2d_valid[:, 0]), min(kp2d_valid[:, 1]),
max(kp2d_valid[:, 0]), max(kp2d_valid[:, 1])]
center = [(bbox[2] + bbox[0]) / 2., (bbox[3] + bbox[1]) / 2.]
scale = self.scale_factor * 1.2 * max(bbox[2] - bbox[0], bbox[3] - bbox[1]) / 200.
res = [constants.IMG_RES, constants.IMG_RES]
ul = np.array(transform([1, 1], center, scale, res, invert=1))-1
# Bottom right point
br = np.array(transform([res[0]+1,
res[1]+1], center, scale, res, invert=1))-1
center = [(ul[0] + br[0]) / 2., (ul[1] + br[1]) / 2.]
width_height = [br[0] - ul[0], br[1] - ul[1]]
bbox = np.array(center + width_height)
bboxes.append(bbox)
scales.append(scale)
self.bboxes = np.stack(bboxes)
self.scales = np.array(scales)
self.image_file_names = self.image_file_names
self.joints2d = joints2d
self.frames = frames
if self.full_body:
joints2d_face = wb_kps['joints2d_face']
joints2d_lhand = wb_kps['joints2d_lhand']
joints2d_rhand = wb_kps['joints2d_rhand']
joints_part = {'lhand': joints2d_lhand, 'rhand': joints2d_rhand, 'face': joints2d_face}
self.bboxes_part = {}
self.joints2d_part = {}
for part, joints in joints_part.items():
# print('joints2d part', part, type(joints), joints[0].shape)
# bboxes, time_pt1, time_pt2 = get_all_bbox_params(joints, vis_thresh=-1)
# bboxes[:, 2:] = 150. / bboxes[:, 2:]
# self.bboxes_part[part] = np.stack([bboxes[:, 0], bboxes[:, 1], bboxes[:, 2], bboxes[:, 2]]).T
# self.joints2d_part[part] = joints[time_pt1:time_pt2]
self.joints2d_part[part] = joints
if len(self.joints2d_part[part]) == 0:
print('part 0000', part, time_pt1, time_pt2, joints[time_pt1:time_pt2])
exit()
def __len__(self):
# return len(self.image_file_names)
return len(self.bboxes)
def rgb_processing(self, rgb_img, center, scale, res, rot=0., flip=0):
"""Process rgb image and do augmentation."""
# in the rgb image we add pixel noise in a channel-wise manner
# rgb_img[:,:,0] = np.minimum(255.0, np.maximum(0.0, rgb_img[:,:,0]*pn[0]))
# rgb_img[:,:,1] = np.minimum(255.0, np.maximum(0.0, rgb_img[:,:,1]*pn[1]))
# rgb_img[:,:,2] = np.minimum(255.0, np.maximum(0.0, rgb_img[:,:,2]*pn[2]))
# crop
crop_img_resized, crop_img, crop_shape = crop(rgb_img, center, scale, res, rot=rot)
# flip the image
if flip:
crop_img_resized = flip_img(crop_img_resized)
crop_img = flip_img(crop_img)
# rgb_img = flip_img(rgb_img)
# (3,224,224),float,[0,1]
crop_img_resized = np.transpose(crop_img_resized.astype('float32'), (2,0,1)) / 255.0
crop_img = np.transpose(crop_img.astype('float32'), (2,0,1)) / 255.0
# rgb_img = np.transpose(rgb_img.astype('float32'), (2,0,1)) / 255.0
return crop_img_resized, crop_img, crop_shape
def j2d_processing(self, kp, t, f, is_smpl=False, is_hand=False, is_face=False, is_feet=False):
"""Process gt 2D keypoints and apply all augmentation transforms."""
kp = kp.copy()
nparts = kp.shape[0]
# res = [constants.IMG_RES, constants.IMG_RES]
# t = get_transform(center, scale, res, rot=rot)
for i in range(nparts):
pt = kp[i,0:2]
new_pt = np.array([pt[0], pt[1], 1.]).T
new_pt = np.dot(t, new_pt)
kp[i,0:2] = new_pt[:2]
# kp[i,0:2] = new_pt[:2].astype(int) + 1
# convert to normalized coordinates
kp[:,:-1] = 2.*kp[:,:-1] / constants.IMG_RES - 1.
# flip the x coordinates
if f:
if is_hand:
kp = flip_kp(kp, type='hand')
elif is_face:
kp = flip_kp(kp, type='face')
elif is_feet:
kp = flip_kp(kp, type='feet')
else:
kp = flip_kp(kp, is_smpl)
kp = kp.astype('float32')
return kp
def __getitem__(self, idx):
if self.pre_load_imgs is not None:
img = self.pre_load_imgs
else:
# img = cv2.cvtColor(cv2.imread(self.image_file_names[idx]), cv2.COLOR_BGR2RGB)
img_orig = cv2.imread(self.image_file_names[idx])[:,:,::-1].copy().astype(np.float32)
# img_orig = img.copy()
orig_height, orig_width = img_orig.shape[:2]
if not self.full_body:
bbox = self.bboxes[idx]
j2d = self.joints2d[idx] if self.has_keypoints else None
norm_img, raw_img, kp_2d = get_single_image_crop_demo(
img,
bbox,
kp_2d=j2d,
scale=self.scale_factor,
crop_size=self.crop_size)
if self.has_keypoints:
return norm_img, kp_2d
else:
return norm_img
else:
item = {}
scale = self.scale_factor
rot = 0.
flip = 0
j2d = self.joints2d[idx]
kp2d_valid = j2d[j2d[:, 2]>0.]
bbox = [min(kp2d_valid[:, 0]), min(kp2d_valid[:, 1]),
max(kp2d_valid[:, 0]), max(kp2d_valid[:, 1])]
center = [(bbox[2] + bbox[0]) / 2., (bbox[3] + bbox[1]) / 2.]
sc = 1.2 * max(bbox[2] - bbox[0], bbox[3] - bbox[1]) / 200.
img, _, crop_shape = self.rgb_processing(img_orig, center, sc*scale, [constants.IMG_RES, constants.IMG_RES])
# crop_img = np.transpose(img.astype('float32'), (1,2,0)) * 255.
# cv2.imwrite('notebooks/output/body_img.png', crop_img.astype(np.uint8))
# Store image before normalization to use it in visualization
item['img_body'] = self.normalize_img(torch.from_numpy(img).float())
item['orig_height'] = orig_height
item['orig_width'] = orig_width
item['person_id'] = self.person_ids[idx]
img_hr, img_crop, _ = self.rgb_processing(img_orig, center, sc*scale, [constants.IMG_RES * 8, constants.IMG_RES * 8])
# print('img_hr', img_hr.shape)
# img_orig = flip_img(img_orig) if flip else img_orig
# img_orig = np.transpose(img_orig.astype('float32'), (2,0,1)) / 255.0
# item['img_orig'] = self.normalize_img(torch.from_numpy(img_orig).float())
kps_transf = get_transform(center, sc * scale, [constants.IMG_RES, constants.IMG_RES], rot=rot)
# rot_transf = get_rot_transf([constants.IMG_RES, constants.IMG_RES], rot)
# item['scale'] = float(sc * scale)
# item['center'] = center.astype(np.float32)
# item['kps_transf'] = get_transform(center, sc * scale, [constants.IMG_RES, constants.IMG_RES], rot=rot).astype(np.float32)
# item['rot_transf'] = rot_transf.astype(np.float32)
lhand_kp2d, rhand_kp2d, face_kp2d = self.joints2d_part['lhand'][idx], self.joints2d_part['rhand'][idx], self.joints2d_part['face'][idx]
hand_kp2d = self.j2d_processing(np.concatenate([lhand_kp2d, rhand_kp2d]).copy(), kps_transf, flip, is_hand=True)
face_kp2d = self.j2d_processing(face_kp2d.copy(), kps_transf, flip, is_face=True)
n_hand_kp = len(constants.HAND_NAMES)
# item['lhand_kp2d'] = hand_kp2d[:n_hand_kp]
# item['rhand_kp2d'] = hand_kp2d[n_hand_kp:]
# item['face_kp2d'] = face_kp2d
# part_kp2d_dict = {'lhand': item['lhand_kp2d'], 'rhand': item['rhand_kp2d'], 'face': item['face_kp2d']}
part_kp2d_dict = {'lhand': hand_kp2d[:n_hand_kp], 'rhand': hand_kp2d[n_hand_kp:], 'face': face_kp2d}
for part in ['lhand', 'rhand', 'face']:
kp2d = part_kp2d_dict[part]
# kp2d_valid = kp2d[kp2d[:, 2]>0.005]
kp2d_valid = kp2d[kp2d[:, 2]>0.]
if len(kp2d_valid) > 0:
bbox = [min(kp2d_valid[:, 0]), min(kp2d_valid[:, 1]),
max(kp2d_valid[:, 0]), max(kp2d_valid[:, 1])]
center_part = [(bbox[2] + bbox[0]) / 2., (bbox[3] + bbox[1]) / 2.]
scale_part = 2. * max(bbox[2] - bbox[0], bbox[3] - bbox[1]) / 2
# handle invalid part keypoints
if len(kp2d_valid) < 1 or scale_part < 0.01:
center_part = [0, 0]
scale_part = 0.5
kp2d[:, 2] = 0
center_part = torch.tensor(center_part).float()
theta_part = torch.zeros(1, 2, 3)
theta_part[:, 0, 0] = scale_part
theta_part[:, 1, 1] = scale_part
theta_part[:, :, -1] = center_part
crop_hf_img_size = torch.Size([1, 3, cfg.MODEL.PyMAF.HF_IMG_SIZE, cfg.MODEL.PyMAF.HF_IMG_SIZE])
grid = F.affine_grid(theta_part.detach(), crop_hf_img_size, align_corners=False)
img_part = F.grid_sample(torch.from_numpy(img_crop[None]), grid.cpu(), align_corners=False).squeeze(0)
item[f'img_{part}'] = self.normalize_img(img_part.float())
theta_i_inv = torch.zeros_like(theta_part)
theta_i_inv[:, 0, 0] = 1. / theta_part[:, 0, 0]
theta_i_inv[:, 1, 1] = 1. / theta_part[:, 1, 1]
theta_i_inv[:, :, -1] = - theta_part[:, :, -1] / theta_part[:, 0, 0].unsqueeze(-1)
# kp2d = torch.from_numpy(kp2d[None])
# part_kp2d = torch.bmm(theta_i_inv, homo_vector(kp2d[:, :, :2]).permute(0, 2, 1)).permute(0, 2, 1)
# part_kp2d = torch.cat([part_kp2d, kp2d[:, :, 2:3]], dim=-1).squeeze(0)
# item[f'{part}_kp2d_local'] = part_kp2d
# item[f'{part}_theta'] = theta_part[0]
item[f'{part}_theta_inv'] = theta_i_inv[0]
return item
# return [item[k] for k in ['img', 'img_lhand', 'img_rhand', 'img_face', 'lhand_theta_inv', 'rhand_theta_inv', 'face_theta_inv']]
class ImageFolder(Dataset):
def __init__(self, image_folder):
self.image_file_names = [
osp.join(image_folder, x)
for x in os.listdir(image_folder)
if x.endswith('.png') or x.endswith('.jpg')
]
self.image_file_names = sorted(self.image_file_names)
def __len__(self):
return len(self.image_file_names)
def __getitem__(self, idx):
img = cv2.cvtColor(cv2.imread(self.image_file_names[idx]), cv2.COLOR_BGR2RGB)
return to_tensor(img)
|