Spaces:
Running
on
Zero
Running
on
Zero
File size: 6,306 Bytes
cf92dec |
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 |
import os
import sys
import traceback
from math import ceil
import PIL.Image
import torch
import distinctipy
import matplotlib.pyplot as plt
from PIL import Image
import numpy as np
import facer
import tyro
from pixel3dmm import env_paths
colors = distinctipy.get_colors(22, rng=0)
def viz_results(img, seq_classes, n_classes, suppress_plot = False):
seg_img = np.zeros([img.shape[-2], img.shape[-1], 3])
#distinctipy.color_swatch(colors)
bad_indices = [
0, # background,
1, # neck
# 2, skin
3, # cloth
4, # ear_r (images-space r)
5, # ear_l
# 6 brow_r
# 7 brow_l
# 8, # eye_r
# 9, # eye_l
# 10 noise
# 11 mouth
# 12 lower_lip
# 13 upper_lip
14, # hair,
# 15, glasses
16, # ??
17, # earring_r
18, # ?
]
bad_indices = []
for i in range(n_classes):
if i not in bad_indices:
seg_img[seq_classes[0, :, :] == i] = np.array(colors[i])*255
if not suppress_plot:
plt.imshow(seg_img.astype(np.uint(8)))
plt.show()
return Image.fromarray(seg_img.astype(np.uint8))
def get_color_seg(img, seq_classes, n_classes):
seg_img = np.zeros([img.shape[-2], img.shape[-1], 3])
colors = distinctipy.get_colors(n_classes+1, rng=0)
#distinctipy.color_swatch(colors)
bad_indices = [
0, # background,
1, # neck
# 2, skin
3, # cloth
4, # ear_r (images-space r)
5, # ear_l
# 6 brow_r
# 7 brow_l
# 8, # eye_r
# 9, # eye_l
# 10 noise
# 11 mouth
# 12 lower_lip
# 13 upper_lip
14, # hair,
# 15, glasses
16, # ??
17, # earring_r
18, # ?
]
for i in range(n_classes):
if i not in bad_indices:
seg_img[seq_classes[0, :, :] == i] = np.array(colors[i])*255
return Image.fromarray(seg_img.astype(np.uint8))
def crop_gt_img(img, seq_classes, n_classes):
seg_img = np.zeros([img.shape[-2], img.shape[-1], 3])
colors = distinctipy.get_colors(n_classes+1, rng=0)
#distinctipy.color_swatch(colors)
bad_indices = [
0, # background,
1, # neck
# 2, skin
3, # cloth
4, #ear_r (images-space r)
5, #ear_l
# 6 brow_r
# 7 brow_l
#8, # eye_r
#9, # eye_l
# 10 noise
# 11 mouth
# 12 lower_lip
# 13 upper_lip
14, # hair,
# 15, glasses
16, # ??
17, # earring_r
18, # ?
]
for i in range(n_classes):
if i in bad_indices:
img[seq_classes[0, :, :] == i] = 0
#plt.imshow(img.astype(np.uint(8)))
#plt.show()
return img.astype(np.uint8)
device = 'cuda' if torch.cuda.is_available() else 'cpu'
face_detector = facer.face_detector('retinaface/mobilenet', device=device)
face_parser = facer.face_parser('farl/celebm/448', device=device) # optional "farl/lapa/448"
def main(video_name : str):
out = f'{env_paths.PREPROCESSED_DATA}/{video_name}'
out_seg = f'{out}/seg_og/'
out_seg_annot = f'{out}/seg_non_crop_annotations/'
os.makedirs(out_seg, exist_ok=True)
os.makedirs(out_seg_annot, exist_ok=True)
folder = f'{out}/cropped/' # '/home/giebenhain/GTA/data_kinect/color/'
frames = [f for f in os.listdir(folder) if f.endswith('.png') or f.endswith('.jpg')]
frames.sort()
if len(os.listdir(out_seg)) == len(frames):
print(f'''
<<<<<<<< ALREADY COMPLETED SEGMENTATION FOR {video_name}, SKIPPING >>>>>>>>
''')
return
#for file in frames:
batch_size = 1
for i in range(len(frames)//batch_size):
image_stack = []
frame_stack = []
original_shapes = []
for j in range(batch_size):
file = frames[i * batch_size + j]
if os.path.exists(f'{out_seg_annot}/color_{file}.png'):
print('DONE')
continue
img = Image.open(f'{folder}/{file}')#.resize((512, 512))
og_size = img.size
image = facer.hwc2bchw(torch.from_numpy(np.array(img)[..., :3])).to(device=device) # image: 1 x 3 x h x w
image_stack.append(image)
frame_stack.append(file[:-4])
for batch_idx in range(ceil(len(image_stack)/batch_size)):
image_batch = torch.cat(image_stack[batch_idx*batch_size:(batch_idx+1)*batch_size], dim=0)
frame_idx_batch = frame_stack[batch_idx*batch_size:(batch_idx+1)*batch_size]
og_shape_batch = original_shapes[batch_idx*batch_size:(batch_idx+1)*batch_size]
#if True:
try:
with torch.inference_mode():
faces = face_detector(image_batch)
torch.cuda.empty_cache()
faces = face_parser(image_batch, faces, bbox_scale_factor=1.25)
torch.cuda.empty_cache()
seg_logits = faces['seg']['logits']
back_ground = torch.all(seg_logits == 0, dim=1, keepdim=True).detach().squeeze(1).cpu().numpy()
seg_probs = seg_logits.softmax(dim=1) # nfaces x nclasses x h x w
seg_classes = seg_probs.argmax(dim=1).detach().cpu().numpy().astype(np.uint8)
seg_classes[back_ground] = seg_probs.shape[1] + 1
for _iidx in range(seg_probs.shape[0]):
frame = frame_idx_batch[_iidx]
iidx = faces['image_ids'][_iidx].item()
try:
I_color = viz_results(image_batch[iidx:iidx+1], seq_classes=seg_classes[_iidx:_iidx+1], n_classes=seg_probs.shape[1] + 1, suppress_plot=True)
I_color.save(f'{out_seg_annot}/color_{frame}.png')
except Exception as ex:
pass
I = Image.fromarray(seg_classes[_iidx])
I.save(f'{out_seg}/{frame}.png')
torch.cuda.empty_cache()
except Exception as exx:
traceback.print_exc()
continue
if __name__ == '__main__':
tyro.cli(main)
|