-
Notifications
You must be signed in to change notification settings - Fork 1
/
Net_deploy.py
406 lines (311 loc) · 14.9 KB
/
Net_deploy.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
# @Time : 11/05/2021
# @Author : Wei Chen
# @Project : Pycharm
from __future__ import print_function
import os
from uti_tool import compute_3d_IoU, loss_recon , show_3D_single
import argparse
import numpy as np
from Net_archs import GCN3D_segR, Rot_green, Rot_red, Point_center_res_cate
import torch
import torch.nn as nn
import cv2
from uti_tool import load_ply, draw_cors_withsize, draw_cors, get_3D_corner, trans_3d, gettrans, get6dpose1
Loss_func_ce = nn.MSELoss()
def load_models(cat, cat1='pd', outf='1', train=False):
classifier_seg3D = GCN3D_segR(class_num=2, vec_num=1, support_num=7, neighbor_num=10)
classifier_ce = Point_center_res_cate() ## translation estimation
classifier_Rot_red = Rot_red(F=1296, k=6) ## rotation red
classifier_Rot_green = Rot_green(F=1296, k=6) ### rotation green
# optimizer = optim.SGD(classifier.parameters(), lr=0.01, momentum=0.9)
classifier_seg3D = nn.DataParallel(classifier_seg3D)
classifier_ce = nn.DataParallel(classifier_ce)
classifier_Rot_red = nn.DataParallel(classifier_Rot_red)
classifier_Rot_green = nn.DataParallel(classifier_Rot_green)
classifier_seg3D = classifier_seg3D.eval()
classifier_ce = classifier_ce.eval()
classifier_Rot_red = classifier_Rot_red.eval()
classifier_Rot_green = classifier_Rot_green.eval()
#
classifier_seg3D.cuda()
classifier_ce.cuda()
classifier_Rot_green.cuda()
classifier_Rot_red.cuda()
if train:
outf = '/home/lcl/fsnet_oper/%s/' % outf # trained models
temp =cat
cat=cat1
Seg3d = '%s/Seg3D_last_obj%s.pth' % (outf, cat)
# cat = temp
Rot = '%s/Rot_g_last_obj%s.pth' % (outf, cat)
Rot_res = '%s/Rot_r_last_obj%s.pth' % (outf, cat)
# cat = 'cb'
Tes = '%s/Tres_last_obj%s.pth' % (outf, cat)
else:
outf = '/home/lcl/fsnet_oper/yolov3_fsnet/trained_models'
temp =cat
cat='3mix_model'
# cat = 'mix_all_ep200'
Seg3d = '%s/Seg3D_last_obj%s.pth' % (outf, cat)
# cat = temp
Rot = '%s/Rot_g_last_obj%s.pth' % (outf, cat)
Rot_res = '%s/Rot_r_last_obj%s.pth' % (outf, cat)
# cat = 'cb'
Tes = '%s/Tres_last_obj%s.pth' % (outf, cat)
cat =temp
classifier_seg3D.load_state_dict(torch.load(Seg3d))
classifier_ce.load_state_dict(torch.load(Tes))
classifier_Rot_green.load_state_dict(torch.load(Rot))
classifier_Rot_red.load_state_dict(torch.load(Rot_res))
# model_sizes = np.array(
# [[120,171,39], [138,129,39], [82,48,67], [75,72,121], [346, 200, 335], [93,74,65], # from grasp
# [27, 62, 81], [48, 33, 96], [92, 94, 29], [93, 74, 65], [25, 47, 88]]) ## 6x3
model_sizes = np.array(
[[120, 171, 39], [138, 129, 39], [82, 48, 67], [60, 58, 97], [346, 200, 335], [93, 74, 65],
[35, 82, 107], [48, 33, 96], [92, 94, 29], [93, 74, 65], [25, 47, 88]]) ## 6x3
cats = ['large_clamp', 'pudding_box', 'potted_meat_can', 'pitcher_base', 'laptop', 'mug',
'cracker_box', 'mustard_bottle', 'power_drill', 'mug_deformed','sugar_box']
cate_id0 = np.where(np.array(cats) == cat)[0][0]
model_size = model_sizes[cate_id0]
return classifier_seg3D, classifier_ce, classifier_Rot_green, classifier_Rot_red, model_size, cate_id0
def load_models_my(cat, model_cat=None):
classifier_seg3D = GCN3D_segR(class_num=2, vec_num=1, support_num=7, neighbor_num=10)
classifier_ce = Point_center_res_cate() ## translation estimation
classifier_Rot_red = Rot_red(F=1296, k=6) ## rotation red
classifier_Rot_green = Rot_green(F=1296, k=6) ### rotation green
# optimizer = optim.SGD(classifier.parameters(), lr=0.01, momentum=0.9)
classifier_seg3D = nn.DataParallel(classifier_seg3D)
classifier_ce = nn.DataParallel(classifier_ce)
classifier_Rot_red = nn.DataParallel(classifier_Rot_red)
classifier_Rot_green = nn.DataParallel(classifier_Rot_green)
classifier_seg3D = classifier_seg3D.eval()
classifier_ce = classifier_ce.eval()
classifier_Rot_red = classifier_Rot_red.eval()
classifier_Rot_green = classifier_Rot_green.eval()
#
classifier_seg3D.cuda()
classifier_ce.cuda()
classifier_Rot_green.cuda()
classifier_Rot_red.cuda()
outf = '/home/potato/workplace/FS_Net-main/models/FS_Net_demo'
temp = cat
cat = 'mix_all'
# cat = 'mix_all_ep200'
Seg3d = '%s/Seg3D_last_obj%s.pth' % (outf, cat)
# cat = temp
Rot = '%s/Rot_g_last_obj%s.pth' % (outf, cat)
Rot_res = '%s/Rot_r_last_obj%s.pth' % (outf, cat)
# cat = 'cb'
Tes = '%s/Tres_last_obj%s.pth' % (outf, cat)
cat = temp
classifier_seg3D.load_state_dict(torch.load(Seg3d))
classifier_ce.load_state_dict(torch.load(Tes))
classifier_Rot_green.load_state_dict(torch.load(Rot))
classifier_Rot_red.load_state_dict(torch.load(Rot_res))
cats = ["banana", "bowl", "box", "can", "cup", "marker", "pear", "sugar"]
#cate_id0 = np.where(np.array(cats) == cat)[0][0]
model_sizes = np.loadtxt("/home/potato/workplace/obj_model/mean_scale_cls.txt")
#model_size = model_sizes[cate_id0]
model_size = model_sizes[0]
cate_id0 = 0
return classifier_seg3D, classifier_ce, classifier_Rot_green, classifier_Rot_red, model_size, cate_id0
def FS_Net_Test(points, pc, rgb, Rt, Tt, classifier_seg3D, classifier_ce, classifier_Rot_green, classifier_Rot_red,
cat, model_size, cate_id0, seg, pts_seg, cen_gt, num_cor=3, pts_rec=0):
OR, x_r, y_r, z_r = get_3D_corner(pc)
points = torch.from_numpy(points.astype(np.float32)).unsqueeze(0)
Rt0 = Rt[0].numpy()
Tt = Tt[0].numpy().reshape(3, 1)
ptsori = points.clone()
points = points.numpy().copy()
res = np.mean(points[0], 0)
points[0, :, 0:3] = points[0, :, 0:3] - np.array([res[0], res[1], res[2]])
points = torch.from_numpy(points).cuda()
pointsf = points[:, :, 0:3].unsqueeze(2) ##128 1500 1 12
points = pointsf.transpose(3, 1)
points_n = pointsf.squeeze(2)
cate_id= torch.zeros((1,1)) # add one hot to GCN
cate_id[0][0]= cate_id0+1
one_hot = torch.zeros(points.shape[0], 16).scatter_(1, cate_id.cpu().long(), 1)
one_hot = one_hot.cuda()
# obj_idh = torch.zeros((1, 1))
#
# if obj_idh.shape[0] == 1:
# obj_idh = obj_idh.view(-1, 1).repeat(points.shape[0], 1)
# else:
# obj_idh = obj_idh.view(-1, 1)
# one_hot = torch.zeros(points.shape[0], 16).scatter_(1, obj_idh.cpu().long(), 1)
#
# one_hot = one_hot.cuda()
model=pc[np.random.choice(len(pc), 500, replace=True),:]
model = torch.from_numpy(model).unsqueeze(0).cuda()
points_n1 = torch.cat([points_n, model], dim=1)
pred_seg, point_recon, feavecs = classifier_seg3D(points_n1, one_hot) # GCN deploy
pts_rec = pts_rec.cuda()
pts_reccp = point_recon
pts_reccp = pts_reccp.cpu()
Ttt=Tt.reshape(1,3)
pts_recon = pts_reccp[0].numpy()+Ttt
show_3D_single(pts_recon,'recon')
loss_vec = loss_recon(point_recon, pts_rec)
# print(' %f' % loss_vec.item())
pred_choice = pred_seg.data.max(2)[1]
Loss_seg3D = nn.CrossEntropyLoss()
# Loss_seg3D.cuda()
loss_seg = Loss_seg3D(pred_seg.reshape(-1, pred_seg.size(-1)).cpu(), seg.view(-1, ).long().cpu())
p = pred_choice
ptsori = ptsori.cuda()
pts_ = torch.index_select(ptsori[0, :, 0:3], 0, p[0, :].nonzero()[:, 0]) ##Nx3 z=500
# seg points from pred recon and visualize
rec_seg = torch.index_select(point_recon[0, :, 0:3], 0, p[0, :].nonzero()[:, 0])
choice_seg = np.random.choice(len(pts_rec), len(rec_seg), replace=True)
pts_rec_seg = pts_rec[choice_seg, :].unsqueeze(0)
reg_seg1 = rec_seg.unsqueeze(0)
loss_vec1 = loss_recon(reg_seg1, pts_rec)
pts_seg_recon = rec_seg.cpu().numpy() + Ttt
show_3D_single(pts_seg_recon, 'recon_seg')
# print('loss rec:%f, loss rec seg:%f' %(loss_vec.item(), loss_vec1.item()))
# choice = np.random.choice(len(dep3d_seg), len(pts_), replace=True)
# dep3d_seg = dep3d_seg[choice, :].astype(np.float32)
# dep3d_seg = torch.tensor(dep3d_seg).unsqueeze(0).cuda()
# pts_1 = pts_.unsqueeze(0).cuda()
# loss_seg = loss_recon(pts_1, dep3d_seg)
feat = torch.index_select(feavecs[0, 0:2000, :], 0, p[0, :].nonzero()[:, 0])
choice1 = np.random.choice(len(feat), 800, replace=True)
choice2 = np.random.choice(500, 200, replace=True)
feat_ori = feat[choice1, :]
feat_model = feavecs[0, 2000:2500, :][choice2, :]
feat = torch.cat([feat_ori, feat_model], dim=0)
if len(pts_) < 10:
print('No object pts')
else:
pts_s = pts_[:, :].unsqueeze(0).float()
# print(ib)
# p[0, 10:31]
# feas = torch.index_select(feass[ib, :, :], 0, indexs[ib, :].nonzero()[:, 0])
if num_cor == 3:
corners0 = torch.Tensor(np.array([[0, 0, 0], [0, 200, 0], [200, 0, 0]]))
else:
corners0 = torch.Tensor(np.array([[0, 0, 0], [0, 200, 0]]))
pts_bak = pts_s.clone()
pts_s = pts_s.cuda()
feat = feat.cuda()
corners0 = corners0.cuda()
pts_s1 = np.asarray(pts_bak[0].cpu())
show_3D_single(pts_s1, 'seg')
# choice = np.random.choice(1500, len(pts_), replace=True)
# pts_seg = pts_seg[:,choice, :]
# pts_s1 = torch.tensor(pts_s1).unsqueeze(0).cuda()
# loss_seg=loss_recon(pts_s1,pts_seg)
pts_s = pts_s.transpose(2, 1)
# pts_s = point_recon.transpose(2, 1)
cate_id0 +=1
cen_pred, obj_size = classifier_ce((pts_s - pts_s.mean(dim=2, keepdim=True)), torch.Tensor([cate_id0]))
# print(cen_pred)
pts_seg_recon =pts_seg_recon.transpose(1,0)
# cen_gt = Tt-np.mean(pts_seg_recon,1).reshape(3,1)
cen_gt = cen_gt[0].cuda()
loss_T=Loss_func_ce(cen_gt,cen_pred)
# print('loss_T:',loss_T.item())
# cen_pred, obj_size = classifier_ce(point_recon.transpose(2, 1), torch.Tensor([cate_id0]))
T_pred = pts_s.mean(dim=2, keepdim=True) + cen_pred.unsqueeze(2) ## 1x3x1
# feavec = torch.cat([box_pred, feat.unsqueeze(0)], 2) ##
feavec = feat.unsqueeze(0).transpose(1, 2)
kp_m = classifier_Rot_green(feavec)
if num_cor == 3:
corners_ = np.array([[0, 0, 0], [0, 1, 0], [1, 0, 0]])
else:
corners_ = np.array([[0, 0, 0], [0, 1, 0]])
kpm_gt = (trans_3d(corners_, Rt0, np.array([0, 0, 0]).T).T).flatten()
bbx_3D = model_size + obj_size.detach().cpu().numpy()
model_3D = np.array([x_r, y_r, z_r])
box_pred_gan = classifier_Rot_red(feat.unsqueeze(0).transpose(1, 2))
pred_axis = np.zeros((num_cor, 3))
pred_axis[0:2, :] = kp_m.view((2, 3)).detach().cpu().numpy()
if num_cor == 3:
pred_axis[2, :] = box_pred_gan.view((2, 3)).detach().cpu().numpy()[1, :]
box_pred_gan = box_pred_gan.detach().cpu().numpy()
box_pred_gan = box_pred_gan / np.linalg.norm(box_pred_gan)
cor0 = corners0.cpu().numpy()
cor0 = cor0 / np.linalg.norm(cor0)
kpm_gt = kpm_gt.reshape((num_cor, 3))
kpm_gt = kpm_gt / np.linalg.norm(kpm_gt)
pred_axis = pred_axis / np.linalg.norm(pred_axis)
pose_gt = gettrans(cor0.reshape((num_cor, 3)), kpm_gt.reshape((num_cor, 1, 3)))
Rt = pose_gt[0][0:3, 0:3]
pose = gettrans(cor0.reshape((num_cor, 3)), pred_axis.reshape((num_cor, 1, 3)))
R = pose[0][0:3, 0:3]
T = (pts_s.mean(dim=2, keepdim=True) + cen_pred.unsqueeze(2)).view(1, 3).detach().cpu().numpy()
# T = res[0:3]+( cen_pred.unsqueeze(2)).view(1, 3).detach().cpu().numpy()
# noise_batch_drop_numofloss_loss__cls_model_epoch.pth
torch.cuda.empty_cache()
show = 1
if show == 1:
R_loss, T_loss = get6dpose1(Rt, Tt, R, T, cat)
# print(Tt,T)
size_2 = bbx_3D.reshape(3)
K = np.array([[591.0125, 0, 322.525], [0, 590.16775, 244.11084], [0, 0, 1]])
rgb0 = rgb
rgb0 = draw_cors(rgb0, pc, K, Rt, Tt, [255, 255, 255])
rgb0 = draw_cors_withsize(rgb0, K, R, T, [255, 0, 0], xr=size_2[0], yr=size_2[1], zr=size_2[2])
font = cv2.FONT_HERSHEY_SIMPLEX
cv2.putText(rgb0, 'R_loss: %s' % (R_loss), (10, 20), font, 0.5, (0, 0, 0), 1, 0)
cv2.putText(rgb0, 'T_loss(mm): %s' % (T_loss), (10, 40), font, 0.5, (0, 0, 0), 1, 0)
cv2.imshow('show', rgb0 / 255)
cv2.waitKey(10)
eva = 1
if eva == 1:
sRT_1 = np.eye(4)
sRT_1[0:3, 0:3] = Rt
sRT_1[0:3, 3:4] = Tt
sRT_2 = np.eye(4)
sRT_2[0:3, 0:3] = R
sRT_2[0:3, 3:4] = T.reshape(3, 1)
size_2 = bbx_3D.reshape(3)
size_1 = model_3D
# size_2 = size_1
class_name_1 = cat
class_name_2 = cat
iou3d = compute_3d_IoU(sRT_1, sRT_2, size_1, size_2, class_name_1, class_name_2,
handle_visibility=1)
return iou3d, R_loss, T_loss, loss_vec.item(), loss_seg.item(), pts_s1, loss_vec1.item(),loss_T.item()
def FS_Net_Test_obj_size(points, pc, classifier_seg3D, classifier_ce, classifier_Rot_green, classifier_Rot_red,
cat, model_size, cate_id0, num_cor=3, pts_rec=0):
points = torch.from_numpy(points.astype(np.float32)).unsqueeze(0)
ptsori = points.clone()
points = points.numpy().copy()
res = np.mean(points[0], 0)
points[0, :, 0:3] = points[0, :, 0:3] - np.array([res[0], res[1], res[2]])
points = torch.from_numpy(points).cuda()
pointsf = points[:, :, 0:3].unsqueeze(2) ##128 1500 1 12
points = pointsf.transpose(3, 1)
points_n = pointsf.squeeze(2)
cate_id = torch.zeros((1, 1)) # add one hot to GCN
cate_id[0][0] = cate_id0 + 1
one_hot = torch.zeros(points.shape[0], 16).scatter_(1, cate_id.cpu().long(), 1)
#one_hot = torch.zeros(points.shape[0], 16).scatter_(1, cate_id.long(), 1)
one_hot = one_hot.cuda()
#print("pc", pc.shape)
model = pc[np.random.choice(len(pc), 500, replace=True), :]
#model = pc
model = torch.from_numpy(model).unsqueeze(0).cuda()
#model = model.cuda()
#print("points", points_n.shape)
#print(model.shape)
points_n1 = torch.cat([points_n.to(model.device), model], dim=1)
pred_seg, point_recon, feavecs = classifier_seg3D(points_n1, one_hot)
pred_choice = pred_seg.data.max(2)[1]
p = pred_choice
ptsori = ptsori.cuda()
#print(" ptsori", ptsori.shape)
ptsori = torch.cat([ptsori, ptsori[:, np.random.choice(ptsori.shape[0], 500, replace=True), :]], dim=1)
pts_ = torch.index_select(ptsori[0, :, 0:3], 0, p[0, :].nonzero()[:, 0]) ##Nx3
if len(pts_) < 10:
print('No object pts')
else:
pts_s = pts_[:, :].unsqueeze(0).float()
pts_s = pts_s.cuda()
pts_s = pts_s.transpose(2, 1)
cate_id0 += 1
cen_pred, obj_size = classifier_ce((pts_s - pts_s.mean(dim=2, keepdim=True)), torch.Tensor([cate_id0]))
bbx_3D = model_size - obj_size.detach().cpu().numpy()
return bbx_3D