/
evaluation.py
333 lines (283 loc) · 17.8 KB
/
evaluation.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
from torch.utils.data import DataLoader
import copy
from progress.bar import Bar
import config
import os
import numpy as np
import torch
import cv2
import matplotlib.pyplot as plt
from utils import torch_op
import util
from RPModule.rpmodule import RelativePoseEstimation,getMatchingPrimitive,RelativePoseEstimation_helper
from RPModule.rputil import opts
import argparse
from model.mymodel import SCNet
import time
from baselines import super4pcs, open3d_global_registration, open3d_fast_global_registration,open3d_color_registration
from open3d import *
import logging
def getLoader(args):
testOption='test'
if 'suncg' in args.dataList:
from datasets.SUNCG import SUNCG as Dataset
dataset_name='suncg'
val_dataset = Dataset(testOption, nViews=config.nViews,meta=False,rotate=False,rgbd=True,hmap=False,segm=True,normal=True,list_=f"./data/dataList/{args.dataList}.npy",singleView=0,entrySplit=args.entrySplit)
elif 'matterport' in args.dataList:
from datasets.Matterport3D import Matterport3D as Dataset
dataset_name='matterport'
val_dataset = Dataset(testOption, nViews=config.nViews,meta=False,rotate=False,rgbd=True,hmap=False,segm=True,normal=True,list_=f"./data/dataList/{args.dataList}.npy",singleView=0,entrySplit=args.entrySplit)
elif 'scannet' in args.dataList:
from datasets.ScanNet import ScanNet as Dataset
dataset_name='scannet'
val_dataset = Dataset(testOption, nViews=config.nViews,meta=False,rotate=False,rgbd=True,hmap=False,segm=True,normal=True,list_=f"./data/dataList/{args.dataList}.npy",singleView=0,fullsize_rgbdn=True,entrySplit=args.entrySplit,representation=args.representation)
if args.debug:
loader = DataLoader(val_dataset, batch_size=1, shuffle=False,drop_last=True,collate_fn=util.collate_fn_cat, worker_init_fn=util.worker_init_fn)
else:
loader = DataLoader(val_dataset, batch_size=1, shuffle=False,num_workers=1,drop_last=True,collate_fn=util.collate_fn_cat, worker_init_fn=util.worker_init_fn)
return dataset_name,loader
def _parse_args():
parser = argparse.ArgumentParser(description='Optional app description')
parser.add_argument('--dataList', type = str, default = 'matterport3dv1', help = 'options: suncgv3,scannetv1,matterport3dv1')
parser.add_argument('--sigmaDist',type=float, default=0.04, help = 'parameter for our pairwise matching algorithm')
parser.add_argument('--sigmaAngle1',type=float, default=0.2615,help = 'parameter for our pairwise matching algorithm')
parser.add_argument('--sigmaAngle2',type=float, default=0.2615, help = 'parameter for our pairwise matching algorithm')
parser.add_argument('--sigmaFeat',type=float, default=0.01, help = 'parameter for our pairwise matching algorithm')
parser.add_argument('--maxIter',type=int,default=1000, help = 'number of pairs to be tested')
parser.add_argument('--outputType',type=str,default='rgbdnsf', help = 'types of output')
parser.add_argument('--debug',action='store_true', help = 'for debug')
parser.add_argument('--exp',type=str,default='', help = 'will create a folder with such name under experiments/')
parser.add_argument('--snumclass',type=int,default=15, help = 'number of semantic class')
parser.add_argument('--featureDim',type=int,default=32, help = 'feature dimension')
parser.add_argument('--maskMethod',type=str,default='second',help='observe the second view')
parser.add_argument('--d',type=str,default='', help = '')
parser.add_argument('--entrySplit',type=int,default=None, help = 'use for parallel eval')
parser.add_argument('--representation',type=str,default='skybox')
parser.add_argument('--method',type=str,choices=['ours','ours_nc','ours_nr','super4pcs','fgs','gs','cgs'],default='ours',help='ours,super4pcs,fgs(fast global registration)')
parser.add_argument('--useTanh', type = int, default = 1, help = 'whether to use tanh layer on feature maps')
parser.add_argument('--saveCompletion', type = int, default = 1, help = 'save the completion result')
parser.add_argument('--batchnorm', type = int, default = 1, help = 'whether to use batch norm in completion network')
parser.add_argument('--skipLayer', type = int, default = 1, help = 'whether to use skil connection in completion network')
parser.add_argument('--num_repeat', type = int, default = 1, help = 'repeat times')
parser.add_argument('--rm',action='store_true',help='will remove previous evaluation named args.exp')
parser.add_argument('--para', type = str, default=None,help = 'file specify parameters for pairwise matching module')
parser.add_argument("-l", "--log", dest="logLevel", choices=['DEBUG', 'INFO', 'WARNING', 'ERROR', 'CRITICAL'], help="Set the logging level")
args = parser.parse_args()
if args.d: os.environ["CUDA_VISIBLE_DEVICES"] = args.d
args.alterStep = 1 if args.method == 'ours_nr' else 3
args.completion = 0 if args.method == 'ours_nc' else 1
args.snumclass = 15 if 'suncg' in args.dataList else 21
if args.logLevel:
logging.basicConfig(level=getattr(logging, args.logLevel))
print("\n parameters... *******************************\n")
print(f"evaluate on {args.dataList}")
print(f"using method: {args.method}")
print(f"mask method: {args.maskMethod}")
if 'ours' in args.method:
print(f"output type: {args.outputType}")
print(f"semantic classes: {args.snumclass}")
print(f"feature dimension: {args.featureDim}")
print(f"skipLayer: {args.skipLayer}")
print("\n parameters... *******************************\n")
time.sleep(5)
args.rpm_para = opts()
args.perStepPara = False
if args.para is not None:
para_val = np.loadtxt(args.para).reshape(-1,4)
args.rpm_para.sigmaAngle1 = para_val[:,0]
args.rpm_para.sigmaAngle2 = para_val[:,1]
args.rpm_para.sigmaDist = para_val[:,2]
args.rpm_para.sigmaFeat = para_val[:,3]
args.perStepPara = True
else:
if args.sigmaAngle1: args.rpm_para.sigmaAngle1 = args.sigmaAngle1
if args.sigmaAngle2: args.rpm_para.sigmaAngle2 = args.sigmaAngle2
if args.sigmaDist: args.rpm_para.sigmaDist = args.sigmaDist
if args.sigmaFeat: args.rpm_para.sigmaFeat = args.sigmaFeat
return args
if __name__ == '__main__':
args = _parse_args()
log = logging.getLogger(__name__)
if not os.path.exists("tmp/rpe"):
os.makedirs("tmp/rpe")
exp_dir = f"tmp/rpe/{args.exp}"
if not os.path.exists(exp_dir):
os.makedirs(exp_dir)
dataset_name,loader = getLoader(args)
bar = Bar('Progress', max=len(loader))
speedBenchmark=[]
Overlaps = ['0-0.1','0.1-0.5','0.5-1.0']
adstatsOverlaps = {it:[] for it in Overlaps}
transstatsOverlaps = {it:[] for it in Overlaps}
error_stats=[]
if not args.rm:
if os.path.exists(f"{exp_dir}/{args.exp}.result.npy"):
error_stats+=np.load(f"{exp_dir}/{args.exp}.result.npy").tolist()
n_run = len(error_stats)//100
args.num_repeat -= n_run
if 'ours' in args.method:
# setup division point of outputs
args.idx_f_start = 3+3+1+args.snumclass
args.idx_f_end = args.idx_f_start + args.featureDim
# initialize network and load checkpoint
net=SCNet(args).cuda()
try:
if 'suncg' in args.dataList:
checkpoint = torch.load('./data/pretrained_model/suncg.comp.pth.tar')
elif 'matterport' in args.dataList:
checkpoint = torch.load('./data/pretrained_model/matterport.comp.pth.tar')
elif 'scannet' in args.dataList:
checkpoint = torch.load('./data/pretrained_model/scannet.comp.pth.tar')
except:
raise Exception("please provide the pretrained model.")
state_dict = checkpoint['state_dict']
net.load_state_dict(state_dict)
net.cuda()
for _ in range(args.num_repeat):
for i, data in enumerate(loader):
st = time.time()
np.random.seed()
# initialize data
rgb,depth,R,Q,norm,imgPath,segm=data['rgb'],data['depth'],data['R'],data['Q'],data['norm'],data['imgsPath'],data['segm']
# use origin size scan for baselines on scannet dataset
if 'scannet' in args.dataList and 'ours' not in args.method:
rgb,depth = data['rgb_full'], data['depth_full']
R = torch_op.npy(R)
rgb = torch_op.npy(rgb*255).clip(0,255).astype('uint8')
norm = torch_op.npy(norm)
depth = torch_op.npy(depth)
segm = torch_op.npy(segm)
R_src = R[0,0,:,:]
R_tgt = R[0,1,:,:]
R_gt_44 = np.matmul(R_tgt,np.linalg.inv(R_src))
R_gt = R_gt_44[:3,:3]
# generate source/target scans, point cloud
depth_src,depth_tgt,normal_src,normal_tgt,color_src,color_tgt,pc_src,pc_tgt = util.parse_data(depth,rgb,norm,args.dataList,args.method)
if len(pc_src) == 0 or len(pc_tgt)==0:
print(f"this point cloud file contain no point")
continue
# compute overlap and other stats
overlap_val,cam_dist_this,pc_dist_this,pc_nn = util.point_cloud_overlap(pc_src, pc_tgt, R_gt_44)
overlap = '0-0.1' if overlap_val <= 0.1 else '0.1-0.5' if overlap_val <= 0.5 else '0.5-1.0'
# do not test non-overlap with traditional method since make no sense.
if args.method in ['fgs','gs','super4pcs','cgs'] and overlap_val < 0.1:
continue
# select which method to evaluate
if args.method == 'super4pcs':
R_hat = super4pcs(pc_src, pc_tgt)
elif args.method == 'fgs':
R_hat = open3d_fast_global_registration(pc_src,pc_tgt)
elif args.method == 'gs':
R_hat = open3d_global_registration(pc_src,pc_tgt)
elif args.method == 'cgs':
R_hat = open3d_color_registration(pc_src,pc_tgt, color_src,color_tgt)
elif 'ours' in args.method:
with torch.set_grad_enabled(False):
data_s = {'rgb': rgb[0,0,:,:,:].transpose(1,2,0),
'depth': depth[0,0,:,:],
'normal':norm[0,0,:,:,:].transpose(1,2,0),
'R': R[0,0,:,:]}
data_t = {'rgb': rgb[0,1,:,:,:].transpose(1,2,0),
'depth': depth[0,1,:,:],
'normal':norm[0,1,:,:,:].transpose(1,2,0),
'R': R[0,1,:,:]}
R_hat = np.eye(4)
# get the complete scans
complete_s=torch.cat((torch_op.v(data['rgb'][:,0,:,:,:]),torch_op.v(data['norm'][:,0,:,:,:]),torch_op.v(data['depth'][:,0:1,:,:])),1)
complete_t=torch.cat((torch_op.v(data['rgb'][:,1,:,:,:]),torch_op.v(data['norm'][:,1,:,:,:]),torch_op.v(data['depth'][:,1:2,:,:])),1)
# apply the observation mask
view_s,mask_s,_ = util.apply_mask(complete_s.clone(),args.maskMethod)
view_t,mask_t,_ = util.apply_mask(complete_t.clone(),args.maskMethod)
mask_s=torch_op.npy(mask_s[0,:,:,:]).transpose(1,2,0)
mask_t=torch_op.npy(mask_t[0,:,:,:]).transpose(1,2,0)
# append mask for valid data
tpmask = (view_s[:,6:7,:,:]!=0).float().cuda()
view_s=torch.cat((view_s,tpmask),1)
tpmask = (view_t[:,6:7,:,:]!=0).float().cuda()
view_t=torch.cat((view_t,tpmask),1)
for alter_ in range(args.alterStep):
# warp the second scan using current transformation estimation
view_t2s=torch_op.v(util.warping(torch_op.npy(view_t),np.linalg.inv(R_hat),args.dataList))
view_s2t=torch_op.v(util.warping(torch_op.npy(view_s),R_hat,args.dataList))
# append the warped scans
view0 = torch.cat((view_s,view_t2s),1)
view1 = torch.cat((view_t,view_s2t),1)
# generate complete scans
f=net(torch.cat((view0,view1)))
f0=f[0:1,:,:,:]
f1=f[1:2,:,:,:]
data_sc,data_tc={},{}
# replace the observed region with observed depth/normal
data_sc['normal'] = (1-mask_s)*torch_op.npy(f0[0,3:6,:,:]).transpose(1,2,0)+mask_s*data_s['normal']
data_tc['normal'] = (1-mask_t)*torch_op.npy(f1[0,3:6,:,:]).transpose(1,2,0)+mask_t*data_t['normal']
data_sc['normal']/= (np.linalg.norm(data_s['normal'],axis=2,keepdims=True)+1e-6)
data_tc['normal']/= (np.linalg.norm(data_t['normal'],axis=2,keepdims=True)+1e-6)
data_sc['depth'] = (1-mask_s[:,:,0])*torch_op.npy(f0[0,6,:,:])+mask_s[:,:,0]*data_s['depth']
data_tc['depth'] = (1-mask_t[:,:,0])*torch_op.npy(f1[0,6,:,:])+mask_t[:,:,0]*data_t['depth']
data_sc['obs_mask'] = mask_s.copy()
data_tc['obs_mask'] = mask_t.copy()
data_sc['rgb'] = (mask_s*data_s['rgb']).astype('uint8')
data_tc['rgb'] = (mask_t*data_t['rgb']).astype('uint8')
# for scannet, we use the original size rgb image(480x640) to extract sift keypoint
if 'scannet' in args.dataList:
data_sc['rgb_full'] = (torch_op.npy(data['rgb_full'][0,0,:,:,:])*255).astype('uint8')
data_tc['rgb_full'] = (torch_op.npy(data['rgb_full'][0,1,:,:,:])*255).astype('uint8')
data_sc['depth_full'] = torch_op.npy(data['depth_full'][0,0,:,:])
data_tc['depth_full'] = torch_op.npy(data['depth_full'][0,1,:,:])
# extract feature maps
f0_feat=f0[:,args.idx_f_start:args.idx_f_end,:,:]
f1_feat=f1[:,args.idx_f_start:args.idx_f_end,:,:]
data_sc['feat']=f0_feat.squeeze(0)
data_tc['feat']=f1_feat.squeeze(0)
# run relative pose module to get next estimate
if args.perStepPara:
para_this = opts(args.rpm_para.sigmaAngle1[alter_],args.rpm_para.sigmaAngle2[alter_],args.rpm_para.sigmaDist[alter_],args.rpm_para.sigmaFeat[alter_])
else:
para_this = args.rpm_para
pts3d,ptt3d,ptsns,ptsnt,dess,dest,ptsW,pttW = getMatchingPrimitive(data_sc,data_tc,dataset_name,args.representation,args.completion)
# early return if too few keypoint detected
if pts3d is None or ptt3d is None or pts3d.shape[0]<2 or pts3d.shape[0]<2:
logging.info(f"no pts detected or less than 2 keypoint detected, return identity: {np.eye(3)}")
R_hat = np.eye(4)
else:
R_hat = RelativePoseEstimation_helper({'pc':pts3d.T,'normal':ptsns,'feat':dess,'weight':ptsW},{'pc':ptt3d.T,'normal':ptsnt,'feat':dest,'weight':pttW},para_this)
# average speed
time_this = time.time()-st
speedBenchmark.append(time_this)
# compute rotation error and translation error
t_hat = R_hat[:3,3]
R_hat = R_hat[:3,:3]
ad_this = util.angular_distance_np(R_hat, R_gt[np.newaxis,:,:])[0]
ad_blind_this = util.angular_distance_np(R_gt[np.newaxis,:,:],np.eye(3)[np.newaxis,:,:])[0]
translation_this = np.linalg.norm(np.matmul((R_hat - R_gt_44[:3,:3]),pc_src.mean(0).reshape(3)) + t_hat - R_gt_44[:3,3])
translation_blind_this = np.linalg.norm(t_hat - R_gt_44[:3,3])
# save result for this pair
R_pred_44=np.eye(4)
R_pred_44[:3,:3]=R_hat
R_pred_44[:3,3]=t_hat
error_stats.append({'img_src':imgPath[0][0],'img_tgt':imgPath[1][0], 'err_ad':ad_this,
'err_t':translation_this,'err_blind':ad_blind_this,'err_t_blind':translation_blind_this,'overlap':overlap_val,'pc_dist':pc_dist_this,
'cam_dist':cam_dist_this,'pc_nearest':pc_nn,'R_gt':R_gt_44,'R_pred_44':R_pred_44})
# update statics
adstatsOverlaps[overlap].append(ad_this)
transstatsOverlaps[overlap].append(translation_this)
# print log
log.info(f"average processing time per pair: {np.sum(speedBenchmark)/len(speedBenchmark)}")
log.info(f"imgPath:{imgPath},R_hat:{R_hat}")
log.info(f"ad/ad_blind this :{ad_this}/{ad_blind_this}\n")
# print progress bar
Bar.suffix = '{dataset:10}: [{0:3}/{1:3}] | Total: {total:} | ETA: {eta:}'.format(i, len(loader), total=bar.elapsed_td, eta=bar.eta_td,dataset=dataset_name)
bar.next()
if (i+1) % 100 == 0:
np.save(f"{exp_dir}/{args.exp}.result.npy",error_stats)
sss=''
for overlap in Overlaps:
sss += f"rotation, overlap:{overlap},nobs:{len(adstatsOverlaps[overlap])}, mean:{np.mean(adstatsOverlaps[overlap])} "
print(sss)
sss=''
for overlap in Overlaps:
sss += f"translation, overlap:{overlap},nobs:{len(transstatsOverlaps[overlap])}, mean:{np.mean(transstatsOverlaps[overlap])} "
print(sss)
if i == args.maxIter:
break
np.save(f"{exp_dir}/{args.exp}.result.npy",error_stats)