-
Notifications
You must be signed in to change notification settings - Fork 1
/
smpl3dclothrec_v7.py
1321 lines (1086 loc) · 46.3 KB
/
smpl3dclothrec_v7.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
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
"""
cloth 3d model reconstruction based on SMPL body model
------------------------------------------------------------
(c) copyright 2019 [email protected]
In : 2D VTON of cloth to SMPL silhouette
SMPL template model params file (pkl)
2D matched cloth image file and mask
Out: plk or npz file for subset of SMPL vertices and displacement vector
Note: the Texture (2D warped cloth) and related 2D vertex and face information is obtained
with original SMPL and camera parameters
For in-advance tesrt purpose of part 3. we could move the pose and apply the displacement vector
we apply the pose and shape params for target user but with same texture and vertices and faces defitnion
template (source: pose and shape) target (pose and shape)
--------------------------------------------------------------------------
SMPL- p smpltemplate.pkl results/viton/smpl/000000.pkl
camera-p smpltemplate.pkl results/viton/smpl/000000.pkl
3D body-v smpl with template param smpl with target params
3D cloth-v displacement obtained use displacemt obtained at template
texture results/viton/2dwarp/00000_1.png same
texture-v cam projected onto the texture same as template (not new vertices)
texture-f model.f same
lightening only for cloth-related vertices same
"""
from __future__ import print_function
import graphutil as graphutil
import boundary_matching
import sys
from os.path import join, exists, abspath, dirname
from os import makedirs
import logging
import cPickle as pickle
import time
import cv2
from PIL import Image
import numpy as np
import chumpy as ch
from opendr.camera import ProjectPoints
from smpl_webuser.serialization import load_model
from smpl_webuser.verts import verts_decorated
from render_model import render_model
import inspect # for debugging
import matplotlib.pyplot as plt
from opendr.lighting import SphericalHarmonics, LambertianPointLight
from opendr.geometry import VertNormals, Rodrigues
from opendr.renderer import TexturedRenderer
import json
from smpl_webuser.lbs import global_rigid_transformation
_LOGGER = logging.getLogger(__name__)
logging.basicConfig(level=logging.INFO)
# To understand and verify the SMPL itself
def _examine_smpl_template(model, detail=False):
print(">> SMPL Template <<<<<<<<<<<<<<<<<<<<<<")
print(type(model))
print(dir(model))
#print('kintree_table', model.kintree_table)
print('pose:', model.pose)
if detail:
print('posedirs:', model.posedirs)
print('betas:', model.betas)
print('shape(model):', model.shape)
if detail:
print('shapedirs:', model.shapedirs)
# print('bs_style:', model.bs_style) # f-m-n
#print('f:', model.f)
print('V template :', type(model.v_template))
print('V template :', model.v_template.shape)
#print('weights:', model.weoptimize_on_jointsights)
print('W type:', type(model.weights))
print('W shape:', model.weights.r.shape)
if detail:
print('W value:')
print(model.weights.r)
#parts = np.count_nonzero(model.weights.r, axis =1)
parts = np.argmax(model.weights.r, axis=1)
print(" :", parts.shape, parts[:6000])
#print('J:', model.J)
#print('v_template:', model.v_template)
#print('J_regressor:', model.J_regressor)
# To understand and verify the paramters
def _examine_smpl_params(params):
print(type(params))
print(params.keys())
print('camera params')
camera = params['cam']
print(" - type:", type(camera))
#print(" - members:", dir(camera))
print(" - cam.t:", camera.t.r) # none-zero, likely only nonzero z
print(" - cam.rt:", camera.rt.r) # zero (fixed)
# print(" - cam.camera_mtx:", camera.camera_mtx) #
print(" - cam.k:", camera.k.r) #
print(" - cam.c:", camera.c.r) #
print(" - cam.f:", camera.f.r) #
# print(params['f'].shape) # 2
print('>> pose')
pose = params['pose']
print("\t\ttype:", type(pose))
print('\t\tshape:', pose.shape) # 72
# convert within
#pose = pose % (2.0*np.pi)
print('\t\tvalues (in degree):')
print(pose*180.0/np.pi) # degree
print('>> betas')
betas = params['betas']
print('\ttype:', type(betas))
print('\tshape:', betas.shape) # 10
# print('\tvalues:', params['betas']) # 10
#
#
#
def build_smplbody_surface(model, pose, betas, cam):
n_betas = betas.shape[0]
viz = False
# 2. build body model
sv = verts_decorated( # surface vertices
trans=ch.zeros(3),
pose=ch.array(pose),
v_template=model.v_template,
J=model.J_regressor,
betas=ch.array(betas),
shapedirs=model.shapedirs[:, :, :n_betas],
weights=model.weights,
kintree_table=model.kintree_table,
bs_style=model.bs_style,
f=model.f,
bs_type=model.bs_type,
posedirs=model.posedirs,
want_Jtr=not viz) # need J_transformed for reposing based on vertices
return sv
#
# 3D Body shape => 2D Body shape => cloth 2D Shape
#
# in order to have the cooresponding body vertex for each cloth vertex, we deform body shape
# here we do the 2D mask based and then later will get the 3D coordinates
#
# model : smpl mesh structure (esp. faces)
# body_sv : 3d body vertices
# cam : camera paramter
# imMask : mask for body + cloth area
# return: clothed2dvt : for body + cloth area
#
def construct_clothed2d_from_body(model, body_sv, j2d, cam, mask):
h, w = mask.shape
# 1. extract edge vertices
# cam should be set with cam.v = body_sv.r
# 1.1 construct face_visibility map in 3D body shape
f_normal, v_normal = graphutil.calc_normal_vectors(cam.v.r, model.f)
face_visibility = graphutil.build_face_visibility(f_normal, cam)
# 1.2. extract edge vertices
check_edge_vertices = True
'''
graph analysis data structure: vertex, edge, vs face
face to edges: mdoel.f
face to vertices: model.f
vertex to edges: graph
vertex to faces: graph
edge to vertices: graph
edge to faces: graph
graph[start_v][end_v][0, 1, 2 (contour label)]
'''
graph, longest_contour_label, con_length = graphutil.build_edge_graph_dict(
cam.v, model.f, face_visibility)
# num_body_vertices = np.count_nonzero(
# graph[:, :, 2] == longest_contour_label)
num_edge_vertices = np.amax(con_length)
#print("edge v number:", num_edge_vertices)
# if save_edge_vertices:
edge_vertices = np.zeros([num_edge_vertices, 2], dtype='int32')
# visualization of contour
img_contour = np.zeros([h, w], dtype='uint8')
i = 0
if check_edge_vertices or save_edge_vertices:
for v_s in range(cam.v.shape[0]):
# for v_e in range(v_s):
for v_e in graph[v_s]:
# if graph[v_s, v_e, 2] == longest_contour_label: # > 0:
if graph[v_s][v_e][2] == longest_contour_label: # > 0:
if check_edge_vertices:
sx, sy = cam.r[v_s] # projected coordinate
ex, ey = cam.r[v_e]
edge_vertices[i, 0], edge_vertices[i, 1] = int(
sx), int(sy)
i = i + 1
cv2.line(img_contour, (int(sx), int(sy)), (int(
ex), int(ey)), graph[v_s][v_e][2], thickness=1)
# boudnary matching
# Body Part
##############################################
# 1.1 read boundary matching input files
#img_idx = 1
#maskfile = "../results/10k/segmentation/10kgt_%04d.png"%img_idx
#mask = cv2.imread(maskfile, cv2.IMREAD_UNCHANGED)
# if mask is None:
# print("cannot open", maskfile), exit()
'''
edge_vertices_path ='edge_vertices_%04d.pkl'%img_idx
with open(edge_vertices_path, 'rb') as f:
edge_vertices = pickle.load(f)
'''
# 1.3 boudnary matching
#neck_xy = (j2d[12,0], j2d[12,1])
neck_y = j2d[12, 1]
lsh_y = j2d[9, 1]
rsh_y = j2d[8, 1]
top_y = int((neck_y + lsh_y + rsh_y)/3.0)
nearest_list, img_allcontours = boundary_matching.boundary_match(
mask, edge_vertices, top_y, step=5)
# print(j2d), print(j2d[12,:2]), exit()
# joints matching added
# j2d will be added for source and tgt ....
njoints = j2d.shape[0]
# 2. transform
# 2.1 adaptation of matching data
nboundarypts = len(nearest_list)
npts = nboundarypts + njoints
srcPts = np.zeros([1, npts, 2], dtype='float32')
tgtPts = np.zeros([1, npts, 2], dtype='float32')
for i in range(nboundarypts):
srcPts[0, i, :] = nearest_list[i][0]
tgtPts[0, i, :] = nearest_list[i][1]
#print(tgtPts[0,i,:], srcPts[0,i,:])
for i in range(nboundarypts, npts):
srcPts[0, i, :] = j2d[i-nboundarypts, :]
tgtPts[0, i, :] = j2d[i-nboundarypts, :]
# 2.2 estimate TPS params
tps = boundary_matching.estimateTPS(srcPts, tgtPts, 10)
# 3. deform the boundary 2D vertices using TPS
'''
body2dvt1 = edge_vertices.astype('float32').reshape(1, -1,2)
print('>>> Edge vertices <<<;-')
print('type:', body2dvt1.dtype)
print('shape:', body2dvt1.shape)
print('x:', np.amin(body2dvt1[:,:,0]), np.amax(body2dvt1[:,:,0]))
print('y:', np.amin(body2dvt1[:,:,1]), np.amax(body2dvt1[:,:,1]))
print(body2dvt1)
transformed = tps.applyTransformation(body2dvt1)
print(transformed)
'''
body2dvt_save = cam.r.copy()
body2dvt = cam.r.copy().reshape(1, -1, 2).astype('float32')
'''
print('>>> all vertices <<<;-')
print('type:', body2dvt.dtype)
print('shape:', body2dvt.shape)
print('x:', np.amin(body2dvt[:,:,0]), np.amax(body2dvt[:,:,0]))
print('y:', np.amin(body2dvt[:,:,1]), np.amax(body2dvt[:,:,1]))
print(body2dvt)
'''
transformed = tps.applyTransformation(body2dvt)
# print(transformed)
clothed2dvt = transformed[1].reshape(-1, 2)
body2dvt = body2dvt.reshape(-1, 2)
return clothed2dvt
def construct_clothed3d_from_clothed2d_depth(body_sv, cam, clothed2d):
# 1. get the dept for body vertex
bodydepth = graphutil.build_depthmap2(body_sv.r, cam)
check_depthmap = False
if check_depthmap:
# depth in reverse way
plt.suptitle('depthmap')
plt.subplot(1, 2, 1)
plt.imshow(img[:, :, ::-1]) # , cmap='gray')
plt.subplot(1, 2, 2)
depthmap = graphutil.build_depthimage(
body_sv.r, model.f, bodydepth, cam, height=h, width=w)
#plt.imshow(depthmap, cmap='gray')
plt.imshow(depthmap)
plt.draw()
plt.show()
# plt.imshow(depthmap, cmap='gray_r') # the closer to camera, the brighter
_ = raw_input('quit?')
#exit()
# 2. modify the depth for clothed
# @TODO
# 3. unproject to 3D
# uv space? pixels coordinated!!
clothuvd = np.zeros(body_sv.r.shape)
clothuvd[:, 0] = clothed2d[:, 0]
clothuvd[:, 1] = clothed2d[:, 1]
# @TODO for now simply use the same depth as body ^^;
clothuvd[:, 2] = bodydepth
cloth3d = cam.unproject_points(clothuvd)
# sv.r = cloth3d # now the model is not body but cloth
return cloth3d
# convert numpy to json for a single person joint
# connvert
# 1) uint8 image to float texture image
# 2) normalize the vertices
# optionally,
# 3) coloring the backsize if face visibiltiy is not None)
# ***note ****: texture coordinate is UP-side Down, and x-y normalized
# j
# def prepare_texture(imv2d, faces, im4texture, skin_color_b, skin_color_g, skin_color_r):
def prepare_texture(imv2d, faces, im4texture):
# add arms to textures
"""img_skin = np.zeros(
[im4texture.shape[0], im4texture.shape[1], 3], dtype=np.uint8)
# img_skin.fill(255) # make skin color
img_skin[:, :, 0] = int(skin_color_b)
img_skin[:, :, 1] = int(skin_color_g)
img_skin[:, :, 2] = int(skin_color_r)
# target cloth + skin-colored painting
im4texture = im4texture + img_skin * (im4texture == 0)"""
# texture = overlayed images of 2d and projected.
texture = im4texture.astype(float)/255.0 # uint8 to float
#print('dtype of img:', img.dtype)
#print('dtype of pjt_texture:', pjt_texture.dtype)
th, tw = texture.shape[0:2]
'''
pjt_texture[:,:,:] = (1.0, .0, .0) #
#pjt_texture[:,:int(tw/2),:] = (1.0, 0., 0.) # B, G, R
pjt_texture[:,int(tw/4):int(3*tw/4),:] = (1.0, 1.0, 1.0) # B, G, R
'''
#print("th, tw:", th, tw)
texture_v2d = np.stack(
(imv2d[:, 0]/tw, imv2d[:, 1]/th), axis=-1) # uv normalize
# 5. project the body model with texture renderer
# 3. reprojection
# print(type(cam.v))
# print(cam.v.r.shape)
#print("textured:", type(pjt_texture), 'dtype:', pjt_texture.dtype, "shape:", pjt_texture.shape)
# print('max:', np.amax(pjt_texture[:, :, 0]), np.amax(
# pjt_texture[:, :, 1]), np.amax(pjt_texture[:, :, 2]))
# print('meam:', np.mean(pjt_texture[:, :, 0]), np.mean(
# pjt_texture[:, :, 1]), np.mean(pjt_texture[:, :, 2]))
# apply the visibility map for texturing
return texture, texture_v2d
def prepare_body_texture(imv2d, c3dw, im_path, im_parse_path):
"""
LIP labels
[(0, 0, 0), # 0=Background
(128, 0, 0), # 1=Hat
(255, 0, 0), # 2=Hair
(0, 85, 0), # 3=Glove
(170, 0, 51), # 4=Sunglasses
(255, 85, 0), # 5=UpperClothes
(0, 0, 85), # 6=Dress
(0, 119, 221), # 7=Coat
(85, 85, 0), # 8=Socks
(0, 85, 85), # 9=Pants
(85, 51, 0), # 10=Jumpsuits
(52, 86, 128), # 11=Scarf
(0, 128, 0), # 12=Skirt
(0, 0, 255), # 13=Face
(51, 170, 221), # 14=LeftArm
(0, 255, 255), # 15=RightArm
(85, 255, 170), # 16=LeftLeg
(170, 255, 85), # 17=RightLeg
(255, 255, 0), # 18=LeftShoe
(255, 170, 0) # 19=RightShoe
(189, 170, 160) # 20=Skin/Neck
]
"""
# get full target body texture except upper-clothes
im = Image.open(im_path)
im_parse = Image.open(im_parse_path)
parse_array = np.array(im_parse)
parse_top = (parse_array == 1) + \
(parse_array == 2) + \
(parse_array == 4) + \
(parse_array == 8) + \
(parse_array == 13) + \
(parse_array == 20)
im_top = im * parse_top - (1 - parse_top) # [-1,1], fill 0 for other parts
parse_cloth = (parse_array == 0) + \
(parse_array == 3) + \
(parse_array == 5) + \
(parse_array == 6) + \
(parse_array == 7) + \
(parse_array == 10) + \
(parse_array == 11) + \
(parse_array == 14) + \
(parse_array == 15) + \
(parse_array == 20)
im_cloth = c3dw * parse_cloth - (1 - parse_cloth) # [-1,1], fill 0 for other parts
parse_bottom = (parse_array == 9) + \
(parse_array == 12) + \
(parse_array == 16) + \
(parse_array == 17) + \
(parse_array == 18) + \
(parse_array == 19) + \
(parse_array == 20)
im_bottom = im * parse_bottom - (1 - parse_bottom) # [-1,1], fill 0 for other parts
# texture = overlayed images of 2d and projected.
full_body = im_top + im_cloth + im_bottom
h, w = full_body.shape[0:2]
h_ext = h * 3//2
im_body_ext = np.zeros([h_ext, w, 3], dtype='uint8')
im_body_ext[:h, :, :] = full_body
# texture = im_body_ext.astype(float)/255.0 # uint8 to float
imHuman = cv2.imread(im_path)
texture = imHuman.astype(float)/255.0 # uint8 to float
th, tw = texture.shape[0:2]
texture_v2d = np.stack(
(imv2d[:, 0]/tw, imv2d[:, 1]/th), axis=-1) # uv normalize
"""plt.subplot(1, 3, 1)
plt.axis('off')
plt.imshow(im_cloth)
plt.title('cloth')
plt.subplot(1, 3, 2)
plt.axis('off')
plt.imshow(im_top)
plt.title('body')
plt.subplot(1, 3, 3)
plt.axis('off')
plt.imshow(full_body)
plt.title('texture')
plt.show()
_ = raw_input("next?")"""
return texture, texture_v2d
#
# texture processing with alpha blending
def prepare_texture_with_alpha(pjt_v, pjt_f, img, mask, target_label):
alpha = np.zeros(mask.shape)
# 1.0 for fully opaque, 0.0 for transparent
alpha[mask == target_label] = 1.0
rgb = img.astype(float)/255.0 # uint8 to float
rgba = cv2.merge((rgb, alpha))
print('shapes:', img.shape, rgb.shape, alpha.shape, rgba.shape)
th, tw = rgba.shape[0:2]
pjt_v[:, 0] = pjt_v[:, 0]/tw # uv normalize
pjt_v[:, 1] = pjt_v[:, 1]/th # uv normalize
return rgba # [:,:,:3]
# create V, A, U, f: geom, bright, cam, renderer
def build_texture_renderer(U, V, f, vt, ft, texture, w, h, ambient=0.0, near=0.5, far=20000, background_image=None):
# add lighting
A = SphericalHarmonics(vn=VertNormals(v=V, f=f),
components=[0., 0., 0., 0., 0., 0., 0., 0., 0.],
light_color=ch.ones(3)) + ambient
"""A = LambertianPointLight(
f=f,
v=V,
num_verts=len(V),
light_pos=ch.array([-500,-500,-500]),
vc=np.ones_like(V.r),
# light_color=ch.array([0.7, 0.7, 0.7])) + 0.3
light_color=ch.array([1.0, 1.0, 1.0])) + 0.5 # brighter"""
if background_image is not None:
R = TexturedRenderer(vc=A, camera=U, f=f, bgcolor=[0.0, 0.0, 0.0],
texture_image=texture, vt=vt, ft=ft,
frustum={'width': w, 'height': h, 'near': near, 'far': far}, background_image=background_image)
else:
R = TexturedRenderer(vc=A, camera=U, f=f, bgcolor=[0.0, 0.0, 0.0],
texture_image=texture, vt=vt, ft=ft,
frustum={'width': w, 'height': h, 'near': near, 'far': far})
return R
# display 3d model
def show_3d_model(cam, _texture, texture_v2d, faces, normalImage=False):
#h, w = imTexture.shape[:2]
h, w = _texture.shape[:2]
dist = 20.0
if normalImage:
texture = prepare_texture(texture_v2d, faces, _texture)
else:
texture = _texture
# 1. build texture renderer
texture_renderer = build_texture_renderer(cam, cam.v, faces, texture_v2d, faces,
texture[::-1, :, :], w, h, 1.0, near=0.5, far=20 + dist)
#textured_cloth2d = texture_renderer.r
# plt.figure()
plt.subplot(1, 5, 1)
plt.axis('off')
plt.imshow(texture[:, :, ::-1])
plt.title('input')
rot_axis = 1
rotation = ch.zeros(3)
rotation[rot_axis] = np.pi/4
img0 = texture_renderer.r[:, :, ::-1]*255.0
img0 = img0.astype('uint8')
for i in range(4):
plt.subplot(1, 5, i+2)
# plt.imshow(pjt_R.r)
plt.imshow(texture_renderer.r)
plt.axis('off')
# plt.draw()
# plt.show()
#plt.title('angle =%f'%yaw)
plt.title('%.0f degree' % (i*45))
cam.v = cam.v.dot(Rodrigues(rotation))
plt.show()
# calcuated the local coordinates at each vetex.
#
# z : normal to the vertex
# x : the smallest indexed neighbor vertex based unit vector
# y : the remianing axis in right handed way, ie. z x x => y
def setup_vertex_local_coord(faces, vertices):
# 1.1 normal vectors (1st axis) at each vertex
_, axis_z = graphutil.calc_normal_vectors(vertices, faces)
# 1.2 get 2nd axis
axis_x = graphutil.find2ndaxis(faces, axis_z, vertices)
# 1.3 get 3rd axis
# matuir contribution. np.cross support row-vectorization
axis_y = np.cross(axis_z[:, :], axis_x[:, :])
return axis_x, axis_y, axis_z
#
# reporesent the displacement (now in global coord) into local coordinates
#
# model: smpl mesh structure
# v0 : reference vertex surface, ie. the body
# v*****array: vertext index array for interest
# d : displacement, ie. v = v0 + d
#
def compute_displacement_at_vertex(model, v0, d_global):
debug = False
# 1.setup local coordinate system to each vertex
axis_x, axis_y, axis_z = setup_vertex_local_coord(model.f, v0)
# 2. express displacement in 3 axises
#dlocal = np.concatenate(np.dot(d, axis_x), np.dot(d, axis_y), np.dot(d, axis_z))
xl = np.sum(d_global*axis_x, axis=1)
yl = np.sum(d_global*axis_y, axis=1)
zl = np.sum(d_global*axis_z, axis=1)
d_local = np.stack((xl, yl, zl), axis=-1)
print('dlocal shape:', xl.shape, yl.shape, zl.shape, d_local.shape)
if debug: # verifying d_global = xs * axis_x + ys* axis_y + z*axis_z
# get global coorindate vector
xg = xl[:, None]*axis_x
yg = yl[:, None]*axis_y
zg = zl[:, None]*axis_z
dg = xg + yg + zg
# check the error
err = np.absolute(dg - d_global)
print('d, e x:', np.amax(d_global[:, 0]), np.amax(
err[:, 0]), np.mean(d_global[:, 0]), np.mean(err[:, 0]))
print('d, e y:', np.amax(d_global[:, 1]), np.amax(
err[:, 1]), np.mean(d_global[:, 1]), np.mean(err[:, 1]))
print('d, e z:', np.amax(d_global[:, 2]), np.amax(
err[:, 2]), np.mean(d_global[:, 2]), np.mean(err[:, 2]))
'''
print('d 0:', np.amax(d_global[:,0]), np.amin(d_global[:,0]))
print('error0:', np.amax(err[:,0]), np.amin(err[:,0]))
print('d 1:', np.amax(d_global[:,1]), np.amin(d_global[:,1]))
print('error1:', np.amax(err[:,1]), np.amin(err[:,1]))
print('d 2:', np.amax(d_global[:,2]), np.amin(d_global[:,2]))
print('error2:', np.amax(err[:,2]), np.amin(err[:,2]))
'''
return d_local
#
# get subset numpy 2-D array of triangles
# whose all 3 vertices or one or them are included in the target vertices set
#
#
def getSubsetFaces(ifaces, set_v, smpl_model, allinclusion):
# get arm/hand vertices
"""set_hand = []
for i in range(smpl_model.shape[0]):
if smpl_model.weights_prior[i][13] > 0 or smpl_model.weights_prior[i][14] > 0 or smpl_model.weights_prior[i][16] > 0 or smpl_model.weights_prior[i][17] > 0 or smpl_model.weights_prior[i][18] > 0 or smpl_model.weights_prior[i][19] > 0 or smpl_model.weights_prior[i][20] > 0 or smpl_model.weights_prior[i][21] > 0 or smpl_model.weights_prior[i][22] > 0 or smpl_model.weights_prior[i][23] > 0:
set_hand.append(i)"""
# get wrist vertices
set_wrist = []
for i in range(smpl_model.shape[0]):
if smpl_model.weights_prior[i][22] > 0 or smpl_model.weights_prior[i][23] > 0:
set_wrist.append(i)
flags = np.zeros(ifaces.shape[0], dtype=np.bool)
for i in range(ifaces.shape[0]):
# no need to check for cloth, we will take all faces
# v1, v2, v3 = ifaces[i]
"""mask_v = np.isin(ifaces[i], set_v)
if (mask_v[0] == True) and (mask_v[1] == True) and (mask_v[2] == True):
flags[i] = True"""
# else:
# flags[i] = False
# add the hand parts
"""hand_v = np.isin(ifaces[i], set_hand)
if (hand_v[0] == True) and (hand_v[1] == True) and (hand_v[2] == True):
flags[i] = True"""
# remove the wrist parts
wrist_v = np.isin(ifaces[i], set_wrist)
if (wrist_v[0] == True) and (wrist_v[1] == True) and (wrist_v[2] == True):
flags[i] = False
else:
flags[i] = True
return ifaces[flags, :]
#
# calculate pixel position of SMPL joints
#
# cam: camera ie. projector
# model: smpl basic mdoel
# sv: surfac vectors (opendr)
# betas : body shape, why needed?
# h: projection image height
# w: projection image width
def calculate_joints(cam, model, sv, betas, h, w):
# 1. get the joint locations
# , 12 ] # index in Jtr # @TODO correct neck
smpl_ids = [8, 5, 2, 1, 4, 7, 21, 19, 17, 16, 18, 20]
# lsh,lelb, lwr, neck
# make the SMPL joints depend on betas
Jdirs = np.dstack([model.J_regressor.dot(model.shapedirs[:, :, i])
for i in range(len(betas))])
J_onbetas = ch.array(Jdirs).dot(betas) + \
model.J_regressor.dot(model.v_template.r)
# get joint positions as a function of model pose, betas and trans
(_, A_global) = global_rigid_transformation(
sv.pose, J_onbetas, model.kintree_table, xp=ch)
Jtr = ch.vstack([g[:3, 3] for g in A_global]) + sv.trans
# add joints, with corresponding to a vertex...
neck_id = 3078 # 2951 #3061 # viton's bewtween shoulder
Jtr = ch.vstack((Jtr, sv[neck_id]))
smpl_ids.append(len(Jtr) - 1)
# head_id = 411
nose_id = 331 # nose vertex id
Jtr = ch.vstack((Jtr, sv[nose_id]))
smpl_ids.append(len(Jtr) - 1)
lear_id = 516
Jtr = ch.vstack((Jtr, sv[lear_id]))
smpl_ids.append(len(Jtr) - 1)
rear_id = 3941 # 422# 226 #396
Jtr = ch.vstack((Jtr, sv[rear_id]))
smpl_ids.append(len(Jtr) - 1)
leye_id = 125 # 220 # 125
Jtr = ch.vstack((Jtr, sv[leye_id]))
smpl_ids.append(len(Jtr) - 1)
reye_id = 3635
Jtr = ch.vstack((Jtr, sv[reye_id]))
smpl_ids.append(len(Jtr) - 1)
# 2. project SMPL joints on the image plane using the estimated camera
cam.v = Jtr
joints_np_wo_confidence = cam.r[smpl_ids] # get the projected value
# print(joints_np_wo_confidence)
joints_np = np.zeros([18, 3])
joints_np[:, :2] = joints_np_wo_confidence
joints_np[:, 2] = 1.0
for i in range(joints_np.shape[0]):
if joints_np[i, 0] < 0 or joints_np[i, 0] > (w-1) or joints_np[i, 1] < 0 or joints_np[i, 1] > (h-1):
joints_np[i, 2] = 0.0
# print(joints_np)
return joints_np
def cvt_joints_np2json(joints_np):
# 1. re-ordering
# same as viton2lsp_joint and reamining
order = [13, 12, 8, 7, 6, 9, 10, 11, 2, 1, 0, 3, 4, 5, 14, 15, 16, 17]
# 2. build dictionary
oneperson = {"face_keypoints": [],
"pose_keypoints": joints_np[order].flatten().tolist(),
"hand_right_keypoints": [],
"hand_left_keypoints": []}
people = [oneperson]
joints_json = {"version": 1.0, "people": people}
return joints_json
def get_skin_color_from_image(im_path, im_parse_path):
imBody = cv2.imread(im_path)
imBodySegm = cv2.imread(im_parse_path)
# neck/skin
skin_color_b = np.mean(imBody[imBodySegm[:, :, 0] == 20, 0])
skin_color_g = np.mean(imBody[imBodySegm[:, :, 1] == 20, 1])
skin_color_r = np.mean(imBody[imBodySegm[:, :, 2] == 20, 2])
print('skin color:', skin_color_r, skin_color_g, skin_color_b)
skin_color = np.mean([skin_color_b, skin_color_g, skin_color_r])
# face
kin_color_b = np.mean(imBody[imBodySegm[:, :, 0] == 13, 0])
kin_color_g = np.mean(imBody[imBodySegm[:, :, 1] == 13, 1])
kin_color_r = np.mean(imBody[imBodySegm[:, :, 2] == 13, 2])
print('face color:', kin_color_r, kin_color_g, kin_color_b)
face_color = np.mean([kin_color_b, kin_color_g, kin_color_r])
# left-arm
in_color_b = np.mean(imBody[imBodySegm[:, :, 0] == 14, 0])
in_color_g = np.mean(imBody[imBodySegm[:, :, 1] == 14, 1])
in_color_r = np.mean(imBody[imBodySegm[:, :, 2] == 14, 2])
print('left-arm color:', in_color_r, in_color_g, in_color_b)
larm_color = np.mean([in_color_b, in_color_g, in_color_r])
# right-arm
n_color_b = np.mean(imBody[imBodySegm[:, :, 0] == 15, 0])
n_color_g = np.mean(imBody[imBodySegm[:, :, 1] == 15, 1])
n_color_r = np.mean(imBody[imBodySegm[:, :, 2] == 15, 2])
print('right-arm color:', n_color_r, n_color_g, n_color_b)
rarm_color = np.mean([n_color_b, n_color_g, n_color_r])
brightest_color = np.max([skin_color, face_color, larm_color, rarm_color])
if face_color == brightest_color:
skin_color_b = kin_color_b
skin_color_g = kin_color_g
skin_color_r = kin_color_r
print("face color is chosen.")
elif larm_color == brightest_color:
skin_color_b = in_color_b
skin_color_g = in_color_g
skin_color_r = in_color_r
print("left-arm color is chosen.")
elif rarm_color == brightest_color:
skin_color_b = n_color_b
skin_color_g = n_color_g
skin_color_r = n_color_r
print("right-arm color is chosen.")
else:
print("skin color is chosen.")
return skin_color_b, skin_color_g, skin_color_r
#
# cloth 3D model reconstrction using 2d cloth (mapped onto template) and template
#
def cloth3drec_core(model, # SMPL model
cam, # camera model, Chv
betas, # shape coef, numpy
n_betas, # num of PCA
pose, # angles, 27x3 numpy
imCloth, # img numpy
imClothMask, # img numpy
human_path, # img numpy
human_segm_path, # img numpy
viz=False): # visualize or not
# get human image and segm
imBody = cv2.imread(human_path)
if imBody is None:
print("cannot open", human_path), exit()
imBodySegm = cv2.imread(human_segm_path, cv2.IMREAD_UNCHANGED)
if imBodySegm is None:
print("cannot open", human_segm_path), exit()
for which in [cam, betas, pose, imCloth, imClothMask, imBody, imBodySegm, model]:
if which is None:
print(retrieve_name(which), 'is None')
exit()
h, w = imCloth.shape[0:2]
h_ext = h * 3//2
print(imClothMask.shape)
print(len(imClothMask.shape))
if len(imClothMask.shape) > 2: # ie. 3 ch to 1 ch
imClothMask = cv2.cvtColor(imClothMask, cv2.COLOR_BGR2GRAY)
# get skin/face color for hand/skin area painting
# skin_color_b, skin_color_g, skin_color_r = get_skin_color_from_image(imBody, imBodySegm)
print(imBodySegm.shape)
print(len(imBodySegm.shape))
if len(imBodySegm.shape) > 2: # ie. 3 ch to 1 ch
imBodySegm = cv2.cvtColor(imBodySegm, cv2.COLOR_BGR2GRAY)
# h vs h_ext
# half body image has size of h x w
# rendering needs full body texture image : size of h_ext x w
'''
# 1. Pose to standard pose
if True: # make standard pose for easier try-on
pose[:] = 0.0
pose[0] = np.pi
# lsh = 16 rsh = 17 67.5 degree rotation around z axis
pose[16*3+2] = -7/16.0*np.pi
pose[17*3+2] = +7/16.0*np.pi
betas[:] = 0.0
#cam.t = [0. , 0., 20.] - cam.t: [ 0. 0. 20.] # [-3.12641449e-03 4.31656201e-01 2.13035413e+01]
cam.t = [0., 0.4, 25.]
cam.rt = [0., 0., 0.]
cam.k = [0., 0., 0., 0., 0.]
cam.f = [5000., 5000.]
cam.c = [ 96., 128.] # depending on the image size
print('Final pose and betas ')
print('pose:', pose.reshape([-1,3]))
print('betas:', betas)
'''
# 1. Prepare input images and masks
# 1.1 build template body model
body_sv = build_smplbody_surface(model, pose, betas, cam)
dist = np.abs(cam.t.r[2] - np.mean(body_sv.r, axis=0)[2])
im3CBlack = np.zeros([h_ext, w, 3], dtype=np.uint8)
imBackground = im3CBlack
imBodyRGB = (render_model(
body_sv.r, model.f, w, h_ext, cam, far=20 + dist, img=imBackground[:, :, ::-1]) * 255.).astype('uint8')
imBodyRGB_body = imBodyRGB.copy()
# 1.2 source cloth image and mask (extension for the same size as body silhouette)
imClothedMask = cv2.cvtColor(
imBodyRGB, cv2.COLOR_BGR2GRAY) # gray silhouette
imClothedMask[imClothedMask > 0] = 255 # binary (0, 1)
# imClothedMask[imClothMask[:,:] > 0] = 255 # union of body and .....
# blank background image
imClothMask_ext = np.zeros([h_ext, w], dtype='uint8')
imClothMask_ext[:h, :] = imClothMask[:, :]
imClothedMask[imClothMask_ext > 0] = 255 # union of body and .....
# blank background image
imCloth_ext = np.zeros([h_ext, w, 3], dtype='uint8')
imCloth_ext[:h, :, :] = imCloth[:, :, :]
imCloth_ext[imClothMask_ext <= 0] = 0 # black out
imBodyRGB[imClothMask_ext > 0] = imCloth_ext[imClothMask_ext > 0]
if viz: # show cloth overlayed on smpl
plt.imshow(imBodyRGB[:h, :, ::-1])
# plt.imshow(imClothedMask)
# print("overlaid")
plt.draw()
plt.show()
_ = raw_input('next?')
# 2. Derive 2D cloth vertics position from body vertices using 2-D deformation
j2d = calculate_joints(cam, model, body_sv, betas, h_ext, w)
j2d_wo_confidence = j2d[:, :2]
cam.v = body_sv
clothed2d = construct_clothed2d_from_body(
model, body_sv, j2d_wo_confidence, cam, imClothedMask)
if viz: # show the clothed 2d vertices
marksize = 1
# blank background image
imClothed2d = np.zeros([h_ext, w], dtype='uint8')
for i in range(clothed2d.shape[0]):
x, y = int(clothed2d[i, 0]), int(clothed2d[i, 1])
imClothed2d[y-marksize:y+marksize, x-marksize:x+marksize] = 255
#cv2.drawMarker(img[:,:, parts[i]], (x,y), 255, markerSize = 5)
plt.imshow(imClothed2d)
plt.draw()
plt.show()
_ = raw_input('next?')
# 3. 3D cloth vertices from 2d position and depth
clothed3d = construct_clothed3d_from_clothed2d_depth(
body_sv, cam, clothed2d)
cam.v = clothed3d # now camera project clothed 3D vertex not body's
# 4. Rendering texture
# 4.1 updatng the cloth mask and image at boundary
'''
imClothMask1d = imClothMask[t].flatten()
print(imClothMask1d.shape)
print(imClothMask1dv4Cloth)
'''
# extend the mask boundary for hide the mismatch beween mask and image for rendering texture
kernel = np.ones((3, 3), np.uint8)
imClothMask_ext_raw = imClothMask_ext.copy()
imClothMask_ext = cv2.dilate(imClothMask_ext, kernel, iterations=2)
# modify the boundary
imClothMask_ext_bndry = imClothMask_ext.copy()
imClothMask_ext_bndry[imClothMask_ext_raw > 0] = 0
imCloth_ext_bndry = imCloth_ext.copy()
imCloth_ext_bndry[:, :, 0] = cv2.dilate(
imCloth_ext[:, :, 0], kernel, iterations=2)
imCloth_ext_bndry[:, :, 1] = cv2.dilate(
imCloth_ext[:, :, 1], kernel, iterations=2)
imCloth_ext_bndry[:, :, 2] = cv2.dilate(
imCloth_ext[:, :, 2], kernel, iterations=2)
# imCloth_ext[imClothMask_ext_bndry > 0, :] = (255, 0, 0) # draw the boundary with Blue
imCloth_ext[imClothMask_ext_bndry > 0,
:] = imCloth_ext_bndry[imClothMask_ext_bndry > 0, :]