forked from canonical-capsules/canonical-capsules
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathconfig.py
406 lines (351 loc) · 15.9 KB
/
config.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
import argparse
# ----------------------------------------
# Global variables within this script
arg_lists = []
parser = argparse.ArgumentParser()
# ----------------------------------------
# Some nice macros to be used for arparse
def str2bool(v):
return v.lower() in ("true", "1")
def add_argument_group(name):
arg = parser.add_argument_group(name)
arg_lists.append(arg)
return arg
# ----------------------------------------
# Arguments for the main program
main_arg = add_argument_group("Main")
main_arg.add_argument("--mode", type=str,
default="train",
choices=["train", "test", "valid", "vis", "eval_regist"],
help="Run mode")
main_arg.add_argument("--data_dump_folder", type=str,
default="data_dump",
help="data_dump_folder saving the data")
main_arg.add_argument("--pc_jitter_type", type=str,
default="None",
help="pc jittter type")
main_arg.add_argument("--dataset", type=str,
default="shapenet",
help="dataset")
main_arg.add_argument("--cat_id", type=int,
default=9,
help="data category")
main_arg.add_argument("--suffix", type=str,
default="",
help="For ease of naming logging folders")
main_arg.add_argument("--cn_type", type=str,
default="acn_b",
help="Encoder context normalization type")
main_arg.add_argument("--pt_file", type=str,
default="",
help="pt file")
main_arg.add_argument("--feat_net", type=str,
default="AcneKpEncoder",
help="Encoder")
# ----------------------------------------
# Arguments for model
model_arg = add_argument_group("model")
model_arg.add_argument("--ae_decoder", type=str,
default="KpDecoder",
help="Decoder to reconstruct point clouds")
model_arg.add_argument("--input_feat", type=str,
default="None",
help="type of input feature")
model_arg.add_argument("--pose_code", type=str,
default="nl-noR_T",
choices=[None, "weighted_qt", "l-localRT", "l-RT", "nl-UStV", "nl-U", "nl-T", "nl-noR_T", "nl-LRF_T", "nl-lq_T"],
help="pose type of capsule")
model_arg.add_argument("--indim", type=int,
default=3,
help="input dimension")
model_arg.add_argument("--num_pts", type=int,
default=1024,
help="num of pts")
model_arg.add_argument("--emb_dims", type=int,
default=512,
help="emb dims")
model_arg.add_argument("--decoder_bottleneck_size", type=int,
default=1280,
help="decoder dims")
model_arg.add_argument("--acne_dim", type=int,
default=128,
help="emb dims for acne")
model_arg.add_argument("--acne_num_g", type=int,
default=10,
help="num_g")
model_arg.add_argument("--acne_net_depth", type=int,
default=3,
help="acne_net_depth")
model_arg.add_argument("--acne_out_dim", type=int,
default=0,
help="acne_out_dim")
model_arg.add_argument("--acne_bn_type", type=str,
default="bn",
help="bn type")
model_arg.add_argument("--bin_score", type=str,
default=None,
help="bin score")
model_arg.add_argument("--mean_type", type=str,
default="q",
help="how to normalize")
model_arg.add_argument("--acn_mean_type", type=str,
default="q",
help="how to normalize")
model_arg.add_argument("--acne_backbone", type=str,
default="None",
help="ablation study with backbone network")
model_arg.add_argument("--acn_reg", type=str2bool,
default=False,
help="whether add regularizer into normalizer")
model_arg.add_argument("--use_pose", type=str2bool,
default=True,
help="use pose")
model_arg.add_argument("--add_noise_kps", type=str2bool,
default=False,
help="noise to kp")
# model_arg.add_argument("--loss_", type=float,
# default=0.0,
# help="add rotation equivariance")
model_arg.add_argument("--loss_align_kp_consistency", type=float,
default=0.0,
help="add rotation equivariance")
model_arg.add_argument("--loss_kps_ref_consistency", type=float,
default=0.0,
help="Ref keypoints should be same.")
model_arg.add_argument("--loss_beta", type=float,
default=0.0,
help="add rotation equivariance")
model_arg.add_argument("--loss_reg_att_f", type=float,
default=0.0,
help="regularizers for the final attention")
model_arg.add_argument("--loss_aligner", type=float,
default=1.0,
help="regularizers for the final attention")
model_arg.add_argument("--loss_equi_assign", type=float,
default=0.0,
help="regularizers for the final attention")
model_arg.add_argument("--loss_equi_r", type=float,
default=0.0,
help="add rotation equivariance")
model_arg.add_argument("--loss_procruste", type=float,
default=0.0,
help="add rotation equivariance")
model_arg.add_argument("--loss_separation", type=float,
default=0.0,
help="add rotation equivariance")
model_arg.add_argument("--loss_2cps_mu", type=float,
default=0.0,
help="add rotation equivariance")
model_arg.add_argument("--loss_ref_kp_can", type=float,
default=0.0,
help="add rotation equivariance")
model_arg.add_argument("--separation_margin", type=float,
default=0.2,
help="separation margin")
model_arg.add_argument("--procruste_noise", type=float,
default=1e-8,
help="separation margin")
model_arg.add_argument("--random_range", type=str,
default="uni-180-0.2",
help="random range")
model_arg.add_argument("--pretrain_pt", type=str,
default="None",
help="norm type for decoder")
model_arg.add_argument("--out_pose_grad", type=str,
default="pc_can",
help="where to multiply pose")
model_arg.add_argument("--decoder_grid", type=str,
default="learnable",
help="type of decoder grid")
model_arg.add_argument("--acne_input_layer", type=str,
default="None",
choices=["conv", "conv_cn_bn_relu", "conv_bn_relu"],
help="type of input layers")
model_arg.add_argument("--acne_num_inner", type=int,
default=2,
help="emb dims for acne")
model_arg.add_argument("--KpDecoderPose", type=str2bool,
default=False,
help="Whether bring points to local")
model_arg.add_argument("--aligner", type=str,
default="init",
help="Whether bring points to local")
model_arg.add_argument("--shift", type=str2bool,
default=False,
help="voting with euclidean")
model_arg.add_argument("--conf_corr", type=str2bool,
default=False,
help="Whether bring points to local")
model_arg.add_argument("--trans_mid", type=str,
default="None",
help="transform the input")
model_arg.add_argument("--pose_block", type=str,
default="None",
help="transform the input")
model_arg.add_argument("--spatial_var_norm", type=str,
default="l1",
help="transform the input")
model_arg.add_argument("--KpDecoderMlp", type=str,
default="nonshare",
help="Whether bring points to local")
model_arg.add_argument("--num_classes", type=int,
default=40,
help="num classes")
model_arg.add_argument("--num_view", type=int,
default=2,
help="num classes")
model_arg.add_argument("--vis_id", type=int,
default=-1,
help="num classes")
model_arg.add_argument("--vis_idx", type=int,
default=-1,
help="num classes")
model_arg.add_argument("--act", type=str,
default="relu",
help="activation")
model_arg.add_argument("--att_type_out", type=str,
default="gmm",
help="attention type of the kout layer")
model_arg.add_argument("--ref_pcd_fn", type=str,
default="None",
help="filename of ref_pcd_fn")
model_arg.add_argument("--patch_pos", type=str,
default="center_att",
help="filename of ref_pcd_fn")
model_arg.add_argument("--ref_kp_type", type=str,
default="None",
help="the way of generating reference kp")
model_arg.add_argument("--num_ref_pcd", type=int,
default=1,
help="num ")
model_arg.add_argument("--num_ref_kp", type=int,
default=0,
help="num ")
model_arg.add_argument("--a_norm_eps", type=str,
default="clamp",
help="how to use compute the a_norm")
# ----------------------------------------
# Arguments for training
train_arg = add_argument_group("Training")
train_arg.add_argument("--model", type=str,
default="AcneAe",
help="model name")
train_arg.add_argument("--scheduler", type=int,
default=1,
help="Adjust learning rate with MultiStepLR")
train_arg.add_argument("--noise_ratio", type=float,
default=0.2,
help="Learning rate (gradient step size)")
train_arg.add_argument("--learning_rate", type=float,
default=1e-3,
help="Learning rate (gradient step size)")
train_arg.add_argument("--weight_decay", type=float,
default=0,
help="Learning rate (gradient step size)")
train_arg.add_argument("--batch_size", type=int,
default=16,
help="Size of each training batch")
train_arg.add_argument("--test_batch_size", type=int,
default=1,
help="Size of each training batch")
train_arg.add_argument("--num_points", type=int,
default=1024,
help="The number of points for point clouds")
train_arg.add_argument("--num_epoch", type=int,
default=450,
help="Number of epochs to train")
train_arg.add_argument("--val_intv", type=int,
default=5000,
help="Validation interval")
train_arg.add_argument("--val_intv_epoch", type=int,
default=10,
help="Validation interval")
train_arg.add_argument("--rep_intv", type=int,
default=100,
help="Report interval")
train_arg.add_argument("--log_dir", type=str,
default="",
help="Directory to save logs")
train_arg.add_argument("--res_dir", type=str,
default="./logs",
help="Directory to save current model")
train_arg.add_argument("--save_dir", type=str,
default="None",
help="Directory to save current model")
train_arg.add_argument("--swap_code", type=str,
default="None",
help="swap code before decoder")
train_arg.add_argument("--pc_align", type=str,
default="x",
help="supervision")
train_arg.add_argument("--resume", type=str2bool,
default=False,
help="Whether to resume training "
"from existing checkpoint")
train_arg.add_argument("--save_feat", type=str2bool,
default=False,
help="Whether to resume training "
"from existing checkpoint")
train_arg.add_argument("--att_chamfer", type=str2bool,
default=False,
help="Whether to resume training "
"from existing checkpoint")
train_arg.add_argument("--rt_grid", type=str2bool,
default=False,
help="Whether to resume training "
"from existing checkpoint")
train_arg.add_argument("--worker", type=int,
default=20,
help="number of workers")
train_arg.add_argument("--grid_dim", type=int,
default=10,
help="number of workers")
train_arg.add_argument("--loss_entropy", type=float,
default=0,
help="loss for entropy")
train_arg.add_argument("--loss_trans_mid", type=float,
default=0,
help="loss for entropy")
train_arg.add_argument("--loss_chamfer_cp", type=float,
default=0,
help="loss for entropy")
train_arg.add_argument("--loss_2cps", type=float,
default=5,
help="loss for entropy")
train_arg.add_argument("--loss_vol_cp", type=float,
default=0,
help="loss for entropy")
train_arg.add_argument("--loss_cov_pts", type=float,
default=0,
help="loss for entropy")
train_arg.add_argument("--loss_att_amount", type=float,
default=1e-3,
help="loss for entropy")
train_arg.add_argument("--loss_orth_att", type=float,
default=0,
help="loss for entropy")
train_arg.add_argument("--loss_chamfer_cp_side", type=float,
default=0,
help="Constraint that center point should be around original points")
train_arg.add_argument("--loss_trans_mid_can", type=float,
default=0,
help="loss for entropy")
train_arg.add_argument("--loss_volume", type=float,
default=1,
help="loss for entropy")
train_arg.add_argument("--loss_reconstruction", type=float,
default=1,
help="loss forreconstruction")
train_arg.add_argument("--loss_decode_cp", type=float,
default=0,
help="loss forreconstruction")
train_arg.add_argument("--use_cuda", type=str2bool,
default=True,
help="cuda seems slower?")
def get_config():
config, unparsed = parser.parse_known_args()
return config, unparsed
def print_usage():
parser.print_usage()
#
# config.py ends here