-
Notifications
You must be signed in to change notification settings - Fork 0
/
Demo.py
431 lines (415 loc) · 25.9 KB
/
Demo.py
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
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
import sys
import os
import time
import json
import numpy as np
import cv2
from pymaf.models import SMPL, pymaf_net
from pymaf.core import path_config
from skimage import measure
import torch
import torch.nn.functional as F
from PIL import Image, ImageOps
from model.util.sdf import *
from model.util.io_util import save_obj_mesh,save_obj_mesh_with_color
from util.obj_io import load_obj_data
import torchvision.transforms as transforms
from cuda_voxelization.cuda_mesh_voxelize import MeshVoxelization,binary_fill_from_corner_3D,SematicVoxelization
from model import ExplicitNet
from config.config import cfg
from model import ImplicitNet
import pdb # pdb.set_trace()
from Constants import consts
from glob import glob
import pickle as pkl
from model.FBNet import define_G
from remove.imutils import process_image
import human_det
from neural_voxelization_layer.smpl_model import TetraSMPL
from neural_voxelization_layer.voxelize import Voxelization
from pytorch3d.structures import Meshes
from pytorch3d.renderer import ( look_at_view_transform, FoVOrthographicCameras, RasterizationSettings, MeshRenderer,
MeshRasterizer, SoftSilhouetteShader)
class MCHumanDemo(object):
def __init__(self):
self.opt = cfg
self.opt.merge_from_file('./config/MCHumanDemo.yaml')
os.environ["CUDA_VISIBLE_DEVICES"] = self.opt.gpu_id
self.cuda =torch.device('cuda')
self.ENet = ExplicitNet(self.opt).to(device=self.cuda)
print('Using Network: ', self.ENet.name)
self.ENet=self.load_weight(self.opt.load_netV_checkpoint_path ,self.ENet)
self.INet = ImplicitNet(self.opt).to(self.cuda)
self.INet = self.load_weight(self.opt.load_netG_checkpoint_path ,self.INet)
print('Using Network: ', self.INet.name)
self.B_MIN = np.array([-consts.real_w/2., -consts.real_h/2., -consts.real_w/2.])
self.B_MAX = np.array([ consts.real_w/2., consts.real_h/2., consts.real_w/2.])
self.nmlFNet=define_G(3, 3, 64, "global", 4, 9, 1, 3, "instance").to(self.cuda)
self.nmlFNet=self.load_weight(self.opt.normal_netF_path ,self.nmlFNet)
self.nmlBNet=define_G(3, 3, 64, "global", 4, 9, 1, 3, "instance").to(self.cuda)
self.nmlBNet=self.load_weight(self.opt.normal_netB_path ,self.nmlBNet)
self.to_tensor = transforms.Compose([
transforms.Resize(self.opt.loadSize),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
])
self.det = human_det.Detection()
self.hps = pymaf_net(path_config.SMPL_MEAN_PARAMS,
pretrained=True).to(self.cuda)
self.hps.load_state_dict(torch.load(
path_config.CHECKPOINT_FILE)['model'],
strict=True)
self.hps.eval()
self.smpl_model = SMPL('./data/smpl/',batch_size=1,create_transl=False).to(self.cuda)
self.faces = torch.Tensor(
self.smpl_model.faces.astype(np.int16)).long().unsqueeze(0).to(
self.cuda)
self.tet_smpl = TetraSMPL('./data/GCMR/basicModel_neutral_lbs_10_207_0_v1.0.0.pkl',
'./data/GCMR/tetra_smpl.npz').to(self.cuda)
self.voxelization = Voxelization(consts.smpl_vertex_code,consts.smpl_tetras ,
sigma=0.05,
smooth_kernel_size=7,
batch_size=1).to(self.cuda)
self.openpose_dir='/home/sunjc0306/openpose-master'
R, T = look_at_view_transform(2, 0, 0)
camera = FoVOrthographicCameras(device=self.cuda,R=R,T=T, scale_xyz=(1 * np.ones(3),))
if True:
raster_settings_silhouette = RasterizationSettings(
image_size=512,
blur_radius=np.log(1. / 1e-4 - 1.) * 5e-5,
faces_per_pixel=50,
cull_backfaces=True,
)
silhouetteRas = MeshRasterizer(
cameras=camera, raster_settings=raster_settings_silhouette)
self.renderer = MeshRenderer(rasterizer=silhouetteRas,
shader=SoftSilhouetteShader())
def load_weight(self,checkpoint_path,model):
if checkpoint_path is not None:
print('loading for net ...', checkpoint_path)
assert (os.path.exists(checkpoint_path))
try:
model.load_state_dict(
self.load_from_multi_GPU(path=checkpoint_path))
except :
model.load_state_dict(
torch.load(checkpoint_path, map_location=self.cuda))
return model
def load_from_multi_GPU(self, path):
# original saved file with DataParallel
state_dict = torch.load(path)
from collections import OrderedDict
new_state_dict = OrderedDict()
for k, v in state_dict.items():
name = k[7:] # remove `module.`
new_state_dict[name] = v
return new_state_dict
def demo(self,img_path,results_path,save_inter=True):
input_data={}
img_icon, img_hps, img_ori, img_mask, uncrop_param = process_image(img_path,self.det, self.opt.loadSize)
img_icon=img_icon.unsqueeze(0).to(self.cuda)
img_mask=img_mask.unsqueeze(0).to(self.cuda)
img_name = img_path.split("/")[-1].rsplit(".", 1)[0]
if save_inter:
img_BGR = ((np.transpose(img_icon[0].detach().cpu().numpy(), (1, 2, 0)) * 0.5 + 0.5) * 255.).astype(np.uint8)[:, :,::-1]
cv2.imwrite('%s/%s_rgb.jpg' % (results_path,img_name), img_BGR)
# return 0.0
img_BGR = ((np.transpose(img_mask[0].detach().cpu().numpy(), (1, 2, 0))) * 255.).astype(np.uint8)[:, :, ::-1]
cv2.imwrite('%s/%s_mask.jpg' % (results_path,img_name), img_BGR)
start_time=time.time()
with torch.no_grad():
normal_F=self.nmlFNet(img_icon)* img_mask
normal_B=self.nmlBNet(img_icon)* img_mask
preds_dict = self.hps(img_hps.to(self.cuda))
output = preds_dict['smpl_out'][-1]
scale, tranX, tranY = output['theta'][0, :3]
data = {}
data['betas'] = output['pred_shape']
data['body_pose'] = output['rotmat'][:, 1:]
data['global_orient'] = output['rotmat'][:, 0:1]
data['smpl_verts'] = output['verts']
trans = torch.tensor([tranX, tranY, 0.0]).to(self.cuda)
data['scale'] = scale
data['trans'] = trans
input_data['img']=torch.cat([img_icon,normal_F,normal_B],dim=1)
if save_inter:
img_BGR = ((np.transpose(normal_F[0].detach().cpu().numpy(), (1, 2, 0)) * 0.5 + 0.5) * 255.).astype(np.uint8)[:, :,::-1]
cv2.imwrite('%s/%s_normal_F.jpg' % (results_path,img_name), img_BGR)
img_BGR = ((np.transpose(normal_B[0].detach().cpu().numpy(), (1, 2, 0)) * 0.5 + 0.5) * 255.).astype(np.uint8)[:, :,::-1]
cv2.imwrite('%s/%s_normal_B.jpg' % (results_path,img_name), img_BGR)
smpl_out = self.smpl_model(betas=data['betas'], body_pose=data['body_pose'], global_orient=data['global_orient'],pose2rot=False)
smpl_verts = (smpl_out.vertices * data['scale']) + data['trans']
smpl_verts *= torch.tensor([1.0, -1.0, -1.0]).to(self.cuda)
save_obj_mesh('%s/%s_init_smpl.obj' % (results_path,img_name), smpl_verts[0].detach().cpu().numpy(),
self.faces[0].detach().cpu().numpy())
img_BGR = ((np.transpose(img_icon[0].detach().cpu().numpy(), (1, 2, 0)) * 0.5 + 0.5) * 255.).astype(np.uint8)[:, :,
::-1]
cv2.imwrite('%s/temp/temp_rgb.jpg' % (os.getcwd()), img_BGR)
cmd = "cd {0}; ./build/examples/openpose/openpose.bin --image_dir {1} --write_json {2}".format(
self.openpose_dir,
os.getcwd() + '/temp',
os.getcwd() + '/temp')
os.system(cmd)
keypoints_path = os.getcwd() + '/temp/' + 'temp_rgb_keypoints.json'
with open(keypoints_path) as fp:
keypointsdata = json.load(fp)
keypoints = []
if 'people' in keypointsdata:
for idx, person_data in enumerate(keypointsdata['people']):
kp_data = np.array(person_data['pose_keypoints_2d'], dtype=np.float32)
kp_data = kp_data.reshape([-1, 3])
kp_data[:, 0] = kp_data[:, 0] * 2 / self.opt.loadSize - 1.0
kp_data[:, 1] = kp_data[:, 1] * 2 / self.opt.loadSize - 1.0
keypoints.append(kp_data)
if len(keypoints) == 0:
keypoints.append(np.zeros([25, 3]))
keypoints = torch.from_numpy(keypoints[0])[None, :, :].float().to(self.cuda)
optimed_betas, optimed_pose, optimed_orient, optimed_trans, smpl_verts=\
self.optm_smpl_param(img_mask[0].permute([1,2,0]),keypoints,data['betas'],data['body_pose'],data['global_orient'],data['scale'],data['trans'],iter_num=100)
with torch.no_grad():
# work_path = '%s/../ICON/%s_smplRefined.pkl' % (results_path, img_name)
# if os.path.exists(work_path):
# print(f'load {work_path}')
# with open(work_path, "rb") as f:
# data_smplRefined = pkl.load(f)
# optimed_orient = torch.from_numpy(data_smplRefined['global_orient'])[None,None].to(self.cuda)
# optimed_pose = torch.from_numpy(data_smplRefined['body_pose'])[None].to(self.cuda)
# optimed_betas = torch.from_numpy(data_smplRefined['betas'])[None].to(self.cuda)
# optimed_trans = torch.from_numpy(data_smplRefined['trans']).to(self.cuda)
# data['scale'] = torch.from_numpy(data_smplRefined['scale']).to(self.cuda)
gt_vert_cam = data['scale'] * self.tet_smpl(torch.cat([optimed_orient,optimed_pose],dim=1), optimed_betas) + optimed_trans
# gt_vert_cam = data['scale'] * self.tet_smpl(torch.cat([data['global_orient'],data['body_pose']],dim=1), data['betas']) + data['trans']
vol = self.voxelization(gt_vert_cam/2)
if save_inter:
save_obj_mesh('%s/%s_optim_smpl.obj' % (results_path,img_name), smpl_verts[0].detach().cpu().numpy(),
self.faces[0].detach().cpu().numpy())
with torch.no_grad():
self.ENet.filter(255*vol)
input_data['deepVoxels'] = self.ENet.im_feat_list[-1] # torch.float32, (B=1,C=8,D=32,H=48,W=32), etc. deepVoxels
if save_inter:
deepVoxels = input_data['deepVoxels'][0].detach().cpu().numpy() # (C=8,D=32,H=48,W=32)
deepVoxels = np.transpose(deepVoxels, (
0, 3, 2, 1)) # (C=8,W=32,H=48,D=32), C-XYZ, np.float32, has both posi/neg values
np.save('%s/%s_deepVoxels.npy' % (results_path,img_name), deepVoxels) # 1.6M
# get est. occu.
if save_inter:
with torch.no_grad():
save_path ='%s/%s_meshCoarse.obj' % (results_path,img_name)
self.ENet.est_occu()
pred_occ = self.ENet.get_preds() # torch.float32, BCDHW, (B,1,128,192,128), est. occupancy
pred_occ = pred_occ[0, 0].detach().cpu().numpy() # DHW
pred_occ = np.transpose(pred_occ, (2, 1, 0)) # (W=128,H=192,D=128), XYZ, np.float32, 0. ~ 1. # WHD, XYZ
verts, faces, normals, _ = measure.marching_cubes_lewiner(pred_occ,level=0.5) # https://scikit-image.org/docs/dev/api/skimage.measure.html#marching-cubes-lewiner
verts = verts * 2.0 # this is only to match the verts_canonization function
# verts = verts_canonization(verts=verts, dim_w=pred_occ.shape[0], dim_h=pred_occ.shape[1])
save_obj_mesh(save_path,verts,faces[:,::-1])
if True:
projection_matrix = np.identity(4)
projection_matrix[0, 0] = 1. / consts.h_normalize_half # const == 2.
projection_matrix[1, 1] = 1. / consts.h_normalize_half # const == 2.
projection_matrix[2, 2] = -1. / consts.h_normalize_half # const == -2., to get inverse depth
calib = torch.Tensor(projection_matrix).float()
input_data['calib'] = calib.unsqueeze(0).to(self.cuda)
with torch.no_grad():
self.INet.filter(input_data['img'])
try:
coords, mat = create_grid(self.opt.resolution_x, self.opt.resolution_y, self.opt.resolution_z,
self.B_MIN, self.B_MAX, transform=None)
def eval_func(points):
points = np.expand_dims(points, axis=0) # (1, 3, num_samples)
points = np.repeat(points, 1, axis=0) # (num_views, 3, num_samples)
samples = torch.from_numpy(points).to(device=self.cuda).float() # (num_views, 3, num_samples)
self.INet.query(points=samples, calibs=input_data['calib'],
deepVoxels=input_data['deepVoxels']) # calib_tensor is (num_views, 4, 4)
pred = self.INet.preds[0][0] # (num_samples,)
return pred.detach().cpu().numpy()
sdf = eval_grid_octree(coords, eval_func, num_samples=self.opt.num_samples)
verts, faces, normals, values = measure.marching_cubes_lewiner(sdf, 0.5)
# transform verts into world coordinate system
verts = np.matmul(mat[:3, :3], verts.T) + mat[:3,3:4] # (3,N), convert verts from voxel-space into mesh-coords
verts = verts.T # (N,3)
obj_result = '%s/%s_meshRefined.obj' % (results_path, img_name)
save_obj_mesh(obj_result, verts*np.array([1.,-1.,-1.]), faces[:,::-1])
verts_tensor = torch.from_numpy(verts.T).unsqueeze(0).to(device=self.cuda).float() # (1, N, 3)
xyz_tensor = self.INet.projection(verts_tensor, input_data['calib'][:1]) # （1, 3, N） Tensor of xyz coordinates in the image plane of (-1,1) zone and of the-first-view
uv = xyz_tensor[:, :2, :] # (1, 2, N) for xy, float -1 ~ 1
color = self.INet.index(img_icon[:1], uv).detach().cpu().numpy()[0].T # (N, 3), RGB, float -1 ~ 1
color = color * 0.5 + 0.5 # (N, 3), RGB, float 0 ~ 1
save_obj_mesh_with_color('%s/%s_color.obj' % (results_path, img_name), verts, faces, color)
# return sdf
except Exception as e:
print(e)
print('Can not create marching cubes at this time.')
return time.time()-start_time
def optm_smpl_param(self,gt_silhouette, keypoint, betas, pose, global_orient, scale, trans, iter_num=0):
# assert iter_num > 0
optimed_pose = torch.tensor(pose, device=self.cuda, requires_grad=True) # [1,23,3,3]
optimed_trans = torch.tensor(trans, device=self.cuda, requires_grad=True) # [3]
optimed_betas = torch.tensor(betas, device=self.cuda, requires_grad=False) # [1,10]
optimed_orient = torch.tensor(global_orient, device=self.cuda, requires_grad=True) # [1,1,3,3]
optimizer_smpl = torch.optim.SGD([optimed_pose, optimed_trans, optimed_betas, optimed_orient], lr=3e-2,
momentum=0.9)
scheduler_smpl = torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer_smpl, mode='min', factor=0.5, verbose=0,
min_lr=1e-5, patience=5)
smpl_out = self.smpl_model(betas=optimed_betas, body_pose=optimed_pose, global_orient=optimed_orient,
pose2rot=False)
smpl_verts = (smpl_out.vertices * scale) + optimed_trans
for i in range(iter_num):
smpl_out = self.smpl_model(betas=optimed_betas, body_pose=optimed_pose, global_orient=optimed_orient,
pose2rot=False)
smpl_verts = (smpl_out.vertices * scale) + optimed_trans
smpl_verts *= torch.tensor([1.0, -1.0, -1.0]).to(self.cuda)
mesh = Meshes(smpl_verts, self.faces).to(self.cuda)
silhouette =self.renderer(mesh)[0, :, :, 3:]
openpose_joints = (smpl_out.joints[:, :25, :] * scale) +optimed_trans
# if i % 10 == 0:
# cv2.imwrite("./debug/%dsilhouette.png" % (i),
# np.uint8((silhouette > 0).detach().cpu().numpy()) * 255)
loss_pose = torch.mean(
(keypoint[:, :, 2:] * openpose_joints[:, :, :2] - keypoint[:, :, 2:] * keypoint[:, :, :2]) ** 2)
diff_S = torch.abs(silhouette - gt_silhouette)
loss_sil = diff_S.mean()
loss =loss_pose* 30.0 + loss_sil * 1.0#+F.l1_loss(optimed_pose,pose)+F.l1_loss(optimed_betas,betas)
optimizer_smpl.zero_grad()
loss.backward()
optimizer_smpl.step()
scheduler_smpl.step(loss)
# print('Iter No.%d: loss = %f, loss_sil = %f, loss_pose = %f' %
# (i, loss.item(), loss_sil.item(), loss_pose.item()))
return optimed_betas, optimed_pose, optimed_orient, optimed_trans, smpl_verts
def demo_with_gt(self,img_path,smpl_path,results_path,save_inter=True):
input_data={}
img_icon, img_hps, img_ori, img_mask, uncrop_param = process_image(img_path,self.det, self.opt.loadSize)
img_icon=img_icon.unsqueeze(0).to(self.cuda)
img_mask=img_mask.unsqueeze(0).to(self.cuda)
img_name = img_path.split("/")[-1].rsplit(".", 1)[0]
if save_inter:
img_BGR = ((np.transpose(img_icon[0].detach().cpu().numpy(), (1, 2, 0)) * 0.5 + 0.5) * 255.).astype(np.uint8)[:, :,::-1]
cv2.imwrite('%s/%s_rgb.jpg' % (results_path,img_name), img_BGR)
# return 0.0
img_BGR = ((np.transpose(img_mask[0].detach().cpu().numpy(), (1, 2, 0))) * 255.).astype(np.uint8)[:, :, ::-1]
cv2.imwrite('%s/%s_mask.jpg' % (results_path,img_name), img_BGR)
start_time=time.time()
with torch.no_grad():
normal_F=self.nmlFNet(img_icon)* img_mask
normal_B=self.nmlBNet(img_icon)* img_mask
input_data['img']=torch.cat([img_icon,normal_F,normal_B],dim=1)
if save_inter:
img_BGR = ((np.transpose(normal_F[0].detach().cpu().numpy(), (1, 2, 0)) * 0.5 + 0.5) * 255.).astype(np.uint8)[:, :,::-1]
cv2.imwrite('%s/%s_normal_F.jpg' % (results_path,img_name), img_BGR)
img_BGR = ((np.transpose(normal_B[0].detach().cpu().numpy(), (1, 2, 0)) * 0.5 + 0.5) * 255.).astype(np.uint8)[:, :,::-1]
cv2.imwrite('%s/%s_normal_B.jpg' % (results_path,img_name), img_BGR)
with torch.no_grad():
if os.path.exists(smpl_path):
with open(smpl_path, "rb") as f:
data_smplRefined = pkl.load(f)
optimed_orient = torch.from_numpy(data_smplRefined['global_orient'])[None,None].to(self.cuda)
optimed_pose = torch.from_numpy(data_smplRefined['body_pose'])[None].to(self.cuda)
optimed_betas = torch.from_numpy(data_smplRefined['betas'])[None].to(self.cuda)
optimed_trans = torch.from_numpy(data_smplRefined['trans']).to(self.cuda)
data['scale'] = torch.from_numpy(data_smplRefined['scale']).to(self.cuda)
gt_vert_cam = data['scale'] * self.tet_smpl(torch.cat([optimed_orient,optimed_pose],dim=1), optimed_betas) + optimed_trans
vol = self.voxelization(gt_vert_cam/2)
if save_inter:
save_obj_mesh('%s/%s_optim_smpl.obj' % (results_path,img_name), smpl_verts[0].detach().cpu().numpy(),
self.faces[0].detach().cpu().numpy())
with torch.no_grad():
self.ENet.filter(255*vol)
input_data['deepVoxels'] = self.ENet.im_feat_list[-1] # torch.float32, (B=1,C=8,D=32,H=48,W=32), etc. deepVoxels
if save_inter:
deepVoxels = input_data['deepVoxels'][0].detach().cpu().numpy() # (C=8,D=32,H=48,W=32)
deepVoxels = np.transpose(deepVoxels, (
0, 3, 2, 1)) # (C=8,W=32,H=48,D=32), C-XYZ, np.float32, has both posi/neg values
np.save('%s/%s_deepVoxels.npy' % (results_path,img_name), deepVoxels) # 1.6M
# get est. occu.
if save_inter:
with torch.no_grad():
save_path ='%s/%s_meshCoarse.obj' % (results_path,img_name)
self.ENet.est_occu()
pred_occ = self.ENet.get_preds() # torch.float32, BCDHW, (B,1,128,192,128), est. occupancy
pred_occ = pred_occ[0, 0].detach().cpu().numpy() # DHW
pred_occ = np.transpose(pred_occ, (2, 1, 0)) # (W=128,H=192,D=128), XYZ, np.float32, 0. ~ 1. # WHD, XYZ
verts, faces, normals, _ = measure.marching_cubes_lewiner(pred_occ,level=0.5) # https://scikit-image.org/docs/dev/api/skimage.measure.html#marching-cubes-lewiner
verts = verts * 2.0 # this is only to match the verts_canonization function
# verts = verts_canonization(verts=verts, dim_w=pred_occ.shape[0], dim_h=pred_occ.shape[1])
save_obj_mesh(save_path,verts,faces[:,::-1])
if True:
projection_matrix = np.identity(4)
projection_matrix[0, 0] = 1. / consts.h_normalize_half # const == 2.
projection_matrix[1, 1] = 1. / consts.h_normalize_half # const == 2.
projection_matrix[2, 2] = -1. / consts.h_normalize_half # const == -2., to get inverse depth
calib = torch.Tensor(projection_matrix).float()
input_data['calib'] = calib.unsqueeze(0).to(self.cuda)
with torch.no_grad():
self.INet.filter(input_data['img'])
try:
coords, mat = create_grid(self.opt.resolution_x, self.opt.resolution_y, self.opt.resolution_z,
self.B_MIN, self.B_MAX, transform=None)
def eval_func(points):
points = np.expand_dims(points, axis=0) # (1, 3, num_samples)
points = np.repeat(points, 1, axis=0) # (num_views, 3, num_samples)
samples = torch.from_numpy(points).to(device=self.cuda).float() # (num_views, 3, num_samples)
self.INet.query(points=samples, calibs=input_data['calib'],
deepVoxels=input_data['deepVoxels']) # calib_tensor is (num_views, 4, 4)
pred = self.INet.preds[0][0] # (num_samples,)
return pred.detach().cpu().numpy()
sdf = eval_grid_octree(coords, eval_func, num_samples=self.opt.num_samples)
verts, faces, normals, values = measure.marching_cubes_lewiner(sdf, 0.5)
# transform verts into world coordinate system
verts = np.matmul(mat[:3, :3], verts.T) + mat[:3,3:4] # (3,N), convert verts from voxel-space into mesh-coords
verts = verts.T # (N,3)
obj_result = '%s/%s_meshRefined.obj' % (results_path, img_name)
save_obj_mesh(obj_result, verts*np.array([1.,-1.,-1.]), faces[:,::-1])
verts_tensor = torch.from_numpy(verts.T).unsqueeze(0).to(device=self.cuda).float() # (1, N, 3)
xyz_tensor = self.INet.projection(verts_tensor, input_data['calib'][:1]) # （1, 3, N） Tensor of xyz coordinates in the image plane of (-1,1) zone and of the-first-view
uv = xyz_tensor[:, :2, :] # (1, 2, N) for xy, float -1 ~ 1
color = self.INet.index(img_icon[:1], uv).detach().cpu().numpy()[0].T # (N, 3), RGB, float -1 ~ 1
color = color * 0.5 + 0.5 # (N, 3), RGB, float 0 ~ 1
save_obj_mesh_with_color('%s/%s_color.obj' % (results_path, img_name), verts, faces, color)
# return sdf
except Exception as e:
print(e)
print('Can not create marching cubes at this time.')
return time.time()-start_time
if __name__ == '__main__':
test=MCHumanDemo()
parser = argparse.ArgumentParser()
parser.add_argument('-input_dir ', '--input_dir', type=str)
parser.add_argument('-SMPL_dir ', '--SMPL_dir ', type=str)
parser.add_argument('-results_path ', '--results_path ', type=str,default='./result')
arg = parser.parse_args()
if arg.SMPL_dir is None:
singTime = test.demo(arg.input_dir,arg.results_path,save_inter=False)
else:
singTime = test.demo_with_gt(arg.input_dir,arg.SMPL_dir,arg.results_path,save_inter=False)
print('The cost of time :',total_tima/count)
# rgbPath='/media/sunjc0306/KESU/dataset/Buff/BuffRender/rgbImage/000000.jpg'
# maskPath='/media/sunjc0306/KESU/dataset/Buff/BuffRender/maskImage/000000.jpg'
# smplPath="/media/sunjc0306/KESU/dataset/Buff/WoGtPaMIR/000000_smpl.obj"
# normal_F_Path='/media/sunjc0306/KESU/dataset/Buff/normal_normal_F/000000.jpg'
# normal_B_Path='/media/sunjc0306/KESU/dataset/Buff/normal_normal_B/000000.jpg'
# img_path_list=glob('/home/sunjc0306/HEI-Human/Pin/*')
# img_path_list.sort(reverse=True)
# count=0
# timeStart=time.time()
# total_tima=0
# results_path='/home/sunjc0306/HEI-Human/Pin'
# for img_path in img_path_list:
# # img=cv2.imread(img_path)
# # img_path="/media/sunjc0306/KESU/dataset/pingterest/ReImage/%6d.jpg"%(count)
# # cv2.imwrite(img_path,img)
# # img_name = img_path.split("/")[-1].rsplit(".", 1)[0]
# # save_path = '%s/%s_meshRefined.obj' % (results_path, img_name)
# # if os.path.exists(save_path):
# # continue
# singTime=test.demo(img_path,results_path,save_inter=False)
# count += 1
# hrsPassed = (time.time() - timeStart) / 3600.
# hrsEachIter = hrsPassed / count
# numItersRemain = len(img_path_list) - count
# hrsRemain = numItersRemain * hrsEachIter # hours that remain
# minsRemain = hrsRemain * 60. # minutes that remain
# img_name = img_path.split("/")[-1].rsplit(".", 1)[0]
# total_tima+=singTime
# print("%s| %.4f | inference: %03d-%03d-%03d | remains %.3f m(s) ......" % (img_name,singTime,0,count,len(img_path_list)-count, minsRemain))
# print('The cost of time :',total_tima/count)