Datasets:

License:
quantaji commited on
Commit
aee200a
1 Parent(s): ceec283

add preprocessing and dataset code

Browse files
configs/alc/semseg-pt-v3m1-0-base-scannet200-debug.py ADDED
@@ -0,0 +1,294 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from pointcept.datasets.preprocessing.scannet.meta_data.scannet200_constants import (
2
+ CLASS_LABELS_200,
3
+ )
4
+
5
+
6
+ _base_ = ["../_base_/default_runtime.py"]
7
+
8
+ # misc custom setting
9
+ batch_size = 2 # bs: total bs in all gpus
10
+ num_worker = 24
11
+ mix_prob = 0.8
12
+ empty_cache = False
13
+ enable_amp = True
14
+
15
+ # model settings
16
+ model = dict(
17
+ type="DefaultSegmentorV2",
18
+ num_classes=200,
19
+ backbone_out_channels=64,
20
+ backbone=dict(
21
+ type="PT-v3m1",
22
+ in_channels=6,
23
+ order=["z", "z-trans", "hilbert", "hilbert-trans"],
24
+ stride=(2, 2, 2, 2),
25
+ enc_depths=(2, 2, 2, 6, 2),
26
+ enc_channels=(32, 64, 128, 256, 512),
27
+ enc_num_head=(2, 4, 8, 16, 32),
28
+ enc_patch_size=(1024, 1024, 1024, 1024, 1024),
29
+ dec_depths=(2, 2, 2, 2),
30
+ dec_channels=(64, 64, 128, 256),
31
+ dec_num_head=(4, 4, 8, 16),
32
+ dec_patch_size=(1024, 1024, 1024, 1024),
33
+ mlp_ratio=4,
34
+ qkv_bias=True,
35
+ qk_scale=None,
36
+ attn_drop=0.0,
37
+ proj_drop=0.0,
38
+ drop_path=0.3,
39
+ shuffle_orders=True,
40
+ pre_norm=True,
41
+ enable_rpe=False,
42
+ enable_flash=True,
43
+ upcast_attention=False,
44
+ upcast_softmax=False,
45
+ cls_mode=False,
46
+ pdnorm_bn=False,
47
+ pdnorm_ln=False,
48
+ pdnorm_decouple=True,
49
+ pdnorm_adaptive=False,
50
+ pdnorm_affine=True,
51
+ pdnorm_conditions=("ScanNet", "S3DIS", "Structured3D"),
52
+ ),
53
+ criteria=[
54
+ dict(type="CrossEntropyLoss", loss_weight=1.0, ignore_index=-1),
55
+ dict(type="LovaszLoss", mode="multiclass", loss_weight=1.0, ignore_index=-1),
56
+ ],
57
+ )
58
+
59
+ # scheduler settings
60
+ epoch = 800
61
+ optimizer = dict(type="AdamW", lr=0.006, weight_decay=0.05)
62
+ scheduler = dict(
63
+ type="OneCycleLR",
64
+ max_lr=[0.006, 0.0006],
65
+ pct_start=0.05,
66
+ anneal_strategy="cos",
67
+ div_factor=10.0,
68
+ final_div_factor=1000.0,
69
+ )
70
+ param_dicts = [dict(keyword="block", lr=0.0006)]
71
+
72
+ # dataset settings
73
+ dataset_type = "ARKitScenesLabelMakerScanNet200Dataset"
74
+ data_root = "data/alc"
75
+
76
+ data = dict(
77
+ num_classes=200,
78
+ ignore_index=-1,
79
+ names=CLASS_LABELS_200,
80
+ train=dict(
81
+ type=dataset_type,
82
+ split="train",
83
+ data_root=data_root,
84
+ transform=[
85
+ dict(type="CenterShift", apply_z=True),
86
+ dict(type="RandomDropout", dropout_ratio=0.2, dropout_application_ratio=0.2),
87
+ # dict(type="RandomRotateTargetAngle", angle=(1/2, 1, 3/2), center=[0, 0, 0], axis="z", p=0.75),
88
+ dict(type="RandomRotate", angle=[-1, 1], axis="z", center=[0, 0, 0], p=0.5),
89
+ dict(type="RandomRotate", angle=[-1 / 64, 1 / 64], axis="x", p=0.5),
90
+ dict(type="RandomRotate", angle=[-1 / 64, 1 / 64], axis="y", p=0.5),
91
+ dict(type="RandomScale", scale=[0.9, 1.1]),
92
+ # dict(type="RandomShift", shift=[0.2, 0.2, 0.2]),
93
+ dict(type="RandomFlip", p=0.5),
94
+ dict(type="RandomJitter", sigma=0.005, clip=0.02),
95
+ dict(type="ElasticDistortion", distortion_params=[[0.2, 0.4], [0.8, 1.6]]),
96
+ dict(type="ChromaticAutoContrast", p=0.2, blend_factor=None),
97
+ dict(type="ChromaticTranslation", p=0.95, ratio=0.05),
98
+ dict(type="ChromaticJitter", p=0.95, std=0.05),
99
+ # dict(type="HueSaturationTranslation", hue_max=0.2, saturation_max=0.2),
100
+ # dict(type="RandomColorDrop", p=0.2, color_augment=0.0),
101
+ dict(
102
+ type="GridSample",
103
+ grid_size=0.02,
104
+ hash_type="fnv",
105
+ mode="train",
106
+ return_grid_coord=True,
107
+ ),
108
+ dict(type="SphereCrop", point_max=102400, mode="random"),
109
+ dict(type="CenterShift", apply_z=False),
110
+ dict(type="NormalizeColor"),
111
+ # dict(type="ShufflePoint"),
112
+ dict(type="ToTensor"),
113
+ dict(
114
+ type="Collect",
115
+ keys=("coord", "grid_coord", "segment"),
116
+ feat_keys=("color", "normal"),
117
+ ),
118
+ ],
119
+ test_mode=False,
120
+ ),
121
+ val=dict(
122
+ type=dataset_type,
123
+ split="val",
124
+ data_root=data_root,
125
+ transform=[
126
+ dict(type="CenterShift", apply_z=True),
127
+ dict(
128
+ type="GridSample",
129
+ grid_size=0.02,
130
+ hash_type="fnv",
131
+ mode="train",
132
+ return_grid_coord=True,
133
+ ),
134
+ dict(type="CenterShift", apply_z=False),
135
+ dict(type="NormalizeColor"),
136
+ dict(type="ToTensor"),
137
+ dict(
138
+ type="Collect",
139
+ keys=("coord", "grid_coord", "segment"),
140
+ feat_keys=("color", "normal"),
141
+ ),
142
+ ],
143
+ test_mode=False,
144
+ ),
145
+ test=dict(
146
+ type=dataset_type,
147
+ split="val",
148
+ data_root=data_root,
149
+ transform=[
150
+ dict(type="CenterShift", apply_z=True),
151
+ dict(type="NormalizeColor"),
152
+ ],
153
+ test_mode=True,
154
+ test_cfg=dict(
155
+ voxelize=dict(
156
+ type="GridSample",
157
+ grid_size=0.02,
158
+ hash_type="fnv",
159
+ mode="test",
160
+ keys=("coord", "color", "normal"),
161
+ return_grid_coord=True,
162
+ ),
163
+ crop=None,
164
+ post_transform=[
165
+ dict(type="CenterShift", apply_z=False),
166
+ dict(type="ToTensor"),
167
+ dict(
168
+ type="Collect",
169
+ keys=("coord", "grid_coord", "index"),
170
+ feat_keys=("color", "normal"),
171
+ ),
172
+ ],
173
+ aug_transform=[
174
+ [
175
+ dict(
176
+ type="RandomRotateTargetAngle",
177
+ angle=[0],
178
+ axis="z",
179
+ center=[0, 0, 0],
180
+ p=1,
181
+ )
182
+ ],
183
+ [
184
+ dict(
185
+ type="RandomRotateTargetAngle",
186
+ angle=[1 / 2],
187
+ axis="z",
188
+ center=[0, 0, 0],
189
+ p=1,
190
+ )
191
+ ],
192
+ [
193
+ dict(
194
+ type="RandomRotateTargetAngle",
195
+ angle=[1],
196
+ axis="z",
197
+ center=[0, 0, 0],
198
+ p=1,
199
+ )
200
+ ],
201
+ [
202
+ dict(
203
+ type="RandomRotateTargetAngle",
204
+ angle=[3 / 2],
205
+ axis="z",
206
+ center=[0, 0, 0],
207
+ p=1,
208
+ )
209
+ ],
210
+ [
211
+ dict(
212
+ type="RandomRotateTargetAngle",
213
+ angle=[0],
214
+ axis="z",
215
+ center=[0, 0, 0],
216
+ p=1,
217
+ ),
218
+ dict(type="RandomScale", scale=[0.95, 0.95]),
219
+ ],
220
+ [
221
+ dict(
222
+ type="RandomRotateTargetAngle",
223
+ angle=[1 / 2],
224
+ axis="z",
225
+ center=[0, 0, 0],
226
+ p=1,
227
+ ),
228
+ dict(type="RandomScale", scale=[0.95, 0.95]),
229
+ ],
230
+ [
231
+ dict(
232
+ type="RandomRotateTargetAngle",
233
+ angle=[1],
234
+ axis="z",
235
+ center=[0, 0, 0],
236
+ p=1,
237
+ ),
238
+ dict(type="RandomScale", scale=[0.95, 0.95]),
239
+ ],
240
+ [
241
+ dict(
242
+ type="RandomRotateTargetAngle",
243
+ angle=[3 / 2],
244
+ axis="z",
245
+ center=[0, 0, 0],
246
+ p=1,
247
+ ),
248
+ dict(type="RandomScale", scale=[0.95, 0.95]),
249
+ ],
250
+ [
251
+ dict(
252
+ type="RandomRotateTargetAngle",
253
+ angle=[0],
254
+ axis="z",
255
+ center=[0, 0, 0],
256
+ p=1,
257
+ ),
258
+ dict(type="RandomScale", scale=[1.05, 1.05]),
259
+ ],
260
+ [
261
+ dict(
262
+ type="RandomRotateTargetAngle",
263
+ angle=[1 / 2],
264
+ axis="z",
265
+ center=[0, 0, 0],
266
+ p=1,
267
+ ),
268
+ dict(type="RandomScale", scale=[1.05, 1.05]),
269
+ ],
270
+ [
271
+ dict(
272
+ type="RandomRotateTargetAngle",
273
+ angle=[1],
274
+ axis="z",
275
+ center=[0, 0, 0],
276
+ p=1,
277
+ ),
278
+ dict(type="RandomScale", scale=[1.05, 1.05]),
279
+ ],
280
+ [
281
+ dict(
282
+ type="RandomRotateTargetAngle",
283
+ angle=[3 / 2],
284
+ axis="z",
285
+ center=[0, 0, 0],
286
+ p=1,
287
+ ),
288
+ dict(type="RandomScale", scale=[1.05, 1.05]),
289
+ ],
290
+ [dict(type="RandomFlip", p=1)],
291
+ ],
292
+ ),
293
+ ),
294
+ )
configs/alc/semseg-pt-v3m1-0-base-scannet200.py ADDED
@@ -0,0 +1,294 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from pointcept.datasets.preprocessing.scannet.meta_data.scannet200_constants import (
2
+ CLASS_LABELS_200,
3
+ )
4
+
5
+
6
+ _base_ = ["../_base_/default_runtime.py"]
7
+
8
+ # misc custom setting
9
+ batch_size = 12 # bs: total bs in all gpus
10
+ num_worker = 24
11
+ mix_prob = 0.8
12
+ empty_cache = False
13
+ enable_amp = True
14
+
15
+ # model settings
16
+ model = dict(
17
+ type="DefaultSegmentorV2",
18
+ num_classes=200,
19
+ backbone_out_channels=64,
20
+ backbone=dict(
21
+ type="PT-v3m1",
22
+ in_channels=6,
23
+ order=["z", "z-trans", "hilbert", "hilbert-trans"],
24
+ stride=(2, 2, 2, 2),
25
+ enc_depths=(2, 2, 2, 6, 2),
26
+ enc_channels=(32, 64, 128, 256, 512),
27
+ enc_num_head=(2, 4, 8, 16, 32),
28
+ enc_patch_size=(1024, 1024, 1024, 1024, 1024),
29
+ dec_depths=(2, 2, 2, 2),
30
+ dec_channels=(64, 64, 128, 256),
31
+ dec_num_head=(4, 4, 8, 16),
32
+ dec_patch_size=(1024, 1024, 1024, 1024),
33
+ mlp_ratio=4,
34
+ qkv_bias=True,
35
+ qk_scale=None,
36
+ attn_drop=0.0,
37
+ proj_drop=0.0,
38
+ drop_path=0.3,
39
+ shuffle_orders=True,
40
+ pre_norm=True,
41
+ enable_rpe=False,
42
+ enable_flash=True,
43
+ upcast_attention=False,
44
+ upcast_softmax=False,
45
+ cls_mode=False,
46
+ pdnorm_bn=False,
47
+ pdnorm_ln=False,
48
+ pdnorm_decouple=True,
49
+ pdnorm_adaptive=False,
50
+ pdnorm_affine=True,
51
+ pdnorm_conditions=("ScanNet", "S3DIS", "Structured3D", "ALC"),
52
+ ),
53
+ criteria=[
54
+ dict(type="CrossEntropyLoss", loss_weight=1.0, ignore_index=-1),
55
+ dict(type="LovaszLoss", mode="multiclass", loss_weight=1.0, ignore_index=-1),
56
+ ],
57
+ )
58
+
59
+ # scheduler settings
60
+ epoch = 800
61
+ optimizer = dict(type="AdamW", lr=0.00161, weight_decay=0.05)
62
+ scheduler = dict(
63
+ type="OneCycleLR",
64
+ max_lr=[0.00161, 0.000161],
65
+ pct_start=0.05,
66
+ anneal_strategy="cos",
67
+ div_factor=10.0,
68
+ final_div_factor=1000.0,
69
+ )
70
+ param_dicts = [dict(keyword="block", lr=0.0006)]
71
+
72
+ # dataset settings
73
+ dataset_type = "ARKitScenesLabelMakerScanNet200Dataset"
74
+ data_root = "data/alc"
75
+
76
+ data = dict(
77
+ num_classes=200,
78
+ ignore_index=-1,
79
+ names=CLASS_LABELS_200,
80
+ train=dict(
81
+ type=dataset_type,
82
+ split="train",
83
+ data_root=data_root,
84
+ transform=[
85
+ dict(type="CenterShift", apply_z=True),
86
+ dict(type="RandomDropout", dropout_ratio=0.2, dropout_application_ratio=0.2),
87
+ # dict(type="RandomRotateTargetAngle", angle=(1/2, 1, 3/2), center=[0, 0, 0], axis="z", p=0.75),
88
+ dict(type="RandomRotate", angle=[-1, 1], axis="z", center=[0, 0, 0], p=0.5),
89
+ dict(type="RandomRotate", angle=[-1 / 64, 1 / 64], axis="x", p=0.5),
90
+ dict(type="RandomRotate", angle=[-1 / 64, 1 / 64], axis="y", p=0.5),
91
+ dict(type="RandomScale", scale=[0.9, 1.1]),
92
+ # dict(type="RandomShift", shift=[0.2, 0.2, 0.2]),
93
+ dict(type="RandomFlip", p=0.5),
94
+ dict(type="RandomJitter", sigma=0.005, clip=0.02),
95
+ dict(type="ElasticDistortion", distortion_params=[[0.2, 0.4], [0.8, 1.6]]),
96
+ dict(type="ChromaticAutoContrast", p=0.2, blend_factor=None),
97
+ dict(type="ChromaticTranslation", p=0.95, ratio=0.05),
98
+ dict(type="ChromaticJitter", p=0.95, std=0.05),
99
+ # dict(type="HueSaturationTranslation", hue_max=0.2, saturation_max=0.2),
100
+ # dict(type="RandomColorDrop", p=0.2, color_augment=0.0),
101
+ dict(
102
+ type="GridSample",
103
+ grid_size=0.02,
104
+ hash_type="fnv",
105
+ mode="train",
106
+ return_grid_coord=True,
107
+ ),
108
+ dict(type="SphereCrop", point_max=102400, mode="random"),
109
+ dict(type="CenterShift", apply_z=False),
110
+ dict(type="NormalizeColor"),
111
+ # dict(type="ShufflePoint"),
112
+ dict(type="ToTensor"),
113
+ dict(
114
+ type="Collect",
115
+ keys=("coord", "grid_coord", "segment"),
116
+ feat_keys=("color", "normal"),
117
+ ),
118
+ ],
119
+ test_mode=False,
120
+ ),
121
+ val=dict(
122
+ type=dataset_type,
123
+ split="val",
124
+ data_root=data_root,
125
+ transform=[
126
+ dict(type="CenterShift", apply_z=True),
127
+ dict(
128
+ type="GridSample",
129
+ grid_size=0.02,
130
+ hash_type="fnv",
131
+ mode="train",
132
+ return_grid_coord=True,
133
+ ),
134
+ dict(type="CenterShift", apply_z=False),
135
+ dict(type="NormalizeColor"),
136
+ dict(type="ToTensor"),
137
+ dict(
138
+ type="Collect",
139
+ keys=("coord", "grid_coord", "segment"),
140
+ feat_keys=("color", "normal"),
141
+ ),
142
+ ],
143
+ test_mode=False,
144
+ ),
145
+ test=dict(
146
+ type=dataset_type,
147
+ split="val",
148
+ data_root=data_root,
149
+ transform=[
150
+ dict(type="CenterShift", apply_z=True),
151
+ dict(type="NormalizeColor"),
152
+ ],
153
+ test_mode=True,
154
+ test_cfg=dict(
155
+ voxelize=dict(
156
+ type="GridSample",
157
+ grid_size=0.02,
158
+ hash_type="fnv",
159
+ mode="test",
160
+ keys=("coord", "color", "normal"),
161
+ return_grid_coord=True,
162
+ ),
163
+ crop=None,
164
+ post_transform=[
165
+ dict(type="CenterShift", apply_z=False),
166
+ dict(type="ToTensor"),
167
+ dict(
168
+ type="Collect",
169
+ keys=("coord", "grid_coord", "index"),
170
+ feat_keys=("color", "normal"),
171
+ ),
172
+ ],
173
+ aug_transform=[
174
+ [
175
+ dict(
176
+ type="RandomRotateTargetAngle",
177
+ angle=[0],
178
+ axis="z",
179
+ center=[0, 0, 0],
180
+ p=1,
181
+ )
182
+ ],
183
+ [
184
+ dict(
185
+ type="RandomRotateTargetAngle",
186
+ angle=[1 / 2],
187
+ axis="z",
188
+ center=[0, 0, 0],
189
+ p=1,
190
+ )
191
+ ],
192
+ [
193
+ dict(
194
+ type="RandomRotateTargetAngle",
195
+ angle=[1],
196
+ axis="z",
197
+ center=[0, 0, 0],
198
+ p=1,
199
+ )
200
+ ],
201
+ [
202
+ dict(
203
+ type="RandomRotateTargetAngle",
204
+ angle=[3 / 2],
205
+ axis="z",
206
+ center=[0, 0, 0],
207
+ p=1,
208
+ )
209
+ ],
210
+ [
211
+ dict(
212
+ type="RandomRotateTargetAngle",
213
+ angle=[0],
214
+ axis="z",
215
+ center=[0, 0, 0],
216
+ p=1,
217
+ ),
218
+ dict(type="RandomScale", scale=[0.95, 0.95]),
219
+ ],
220
+ [
221
+ dict(
222
+ type="RandomRotateTargetAngle",
223
+ angle=[1 / 2],
224
+ axis="z",
225
+ center=[0, 0, 0],
226
+ p=1,
227
+ ),
228
+ dict(type="RandomScale", scale=[0.95, 0.95]),
229
+ ],
230
+ [
231
+ dict(
232
+ type="RandomRotateTargetAngle",
233
+ angle=[1],
234
+ axis="z",
235
+ center=[0, 0, 0],
236
+ p=1,
237
+ ),
238
+ dict(type="RandomScale", scale=[0.95, 0.95]),
239
+ ],
240
+ [
241
+ dict(
242
+ type="RandomRotateTargetAngle",
243
+ angle=[3 / 2],
244
+ axis="z",
245
+ center=[0, 0, 0],
246
+ p=1,
247
+ ),
248
+ dict(type="RandomScale", scale=[0.95, 0.95]),
249
+ ],
250
+ [
251
+ dict(
252
+ type="RandomRotateTargetAngle",
253
+ angle=[0],
254
+ axis="z",
255
+ center=[0, 0, 0],
256
+ p=1,
257
+ ),
258
+ dict(type="RandomScale", scale=[1.05, 1.05]),
259
+ ],
260
+ [
261
+ dict(
262
+ type="RandomRotateTargetAngle",
263
+ angle=[1 / 2],
264
+ axis="z",
265
+ center=[0, 0, 0],
266
+ p=1,
267
+ ),
268
+ dict(type="RandomScale", scale=[1.05, 1.05]),
269
+ ],
270
+ [
271
+ dict(
272
+ type="RandomRotateTargetAngle",
273
+ angle=[1],
274
+ axis="z",
275
+ center=[0, 0, 0],
276
+ p=1,
277
+ ),
278
+ dict(type="RandomScale", scale=[1.05, 1.05]),
279
+ ],
280
+ [
281
+ dict(
282
+ type="RandomRotateTargetAngle",
283
+ angle=[3 / 2],
284
+ axis="z",
285
+ center=[0, 0, 0],
286
+ p=1,
287
+ ),
288
+ dict(type="RandomScale", scale=[1.05, 1.05]),
289
+ ],
290
+ [dict(type="RandomFlip", p=1)],
291
+ ],
292
+ ),
293
+ ),
294
+ )
configs/alc/semseg-pt-v3m1-0-base-wn199-debug.py ADDED
@@ -0,0 +1,291 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from pointcept.datasets.preprocessing.alc.preprocess_arkitscenes_labelmaker_consensus import WORDNET_NAMES
2
+
3
+ _base_ = ["../_base_/default_runtime.py"]
4
+
5
+ # misc custom setting
6
+ batch_size = 1 # bs: total bs in all gpus
7
+ num_worker = 24
8
+ mix_prob = 0.8
9
+ empty_cache = False
10
+ enable_amp = True
11
+
12
+ # model settings
13
+ model = dict(
14
+ type="DefaultSegmentorV2",
15
+ num_classes=185,
16
+ backbone_out_channels=64,
17
+ backbone=dict(
18
+ type="PT-v3m1",
19
+ in_channels=6,
20
+ order=["z", "z-trans", "hilbert", "hilbert-trans"],
21
+ stride=(2, 2, 2, 2),
22
+ enc_depths=(2, 2, 2, 6, 2),
23
+ enc_channels=(32, 64, 128, 256, 512),
24
+ enc_num_head=(2, 4, 8, 16, 32),
25
+ enc_patch_size=(1024, 1024, 1024, 1024, 1024),
26
+ dec_depths=(2, 2, 2, 2),
27
+ dec_channels=(64, 64, 128, 256),
28
+ dec_num_head=(4, 4, 8, 16),
29
+ dec_patch_size=(1024, 1024, 1024, 1024),
30
+ mlp_ratio=4,
31
+ qkv_bias=True,
32
+ qk_scale=None,
33
+ attn_drop=0.0,
34
+ proj_drop=0.0,
35
+ drop_path=0.3,
36
+ shuffle_orders=True,
37
+ pre_norm=True,
38
+ enable_rpe=False,
39
+ enable_flash=True,
40
+ upcast_attention=False,
41
+ upcast_softmax=False,
42
+ cls_mode=False,
43
+ pdnorm_bn=False,
44
+ pdnorm_ln=False,
45
+ pdnorm_decouple=True,
46
+ pdnorm_adaptive=False,
47
+ pdnorm_affine=True,
48
+ pdnorm_conditions=("ScanNet", "S3DIS", "Structured3D"),
49
+ ),
50
+ criteria=[
51
+ dict(type="CrossEntropyLoss", loss_weight=1.0, ignore_index=-1),
52
+ dict(type="LovaszLoss", mode="multiclass", loss_weight=1.0, ignore_index=-1),
53
+ ],
54
+ )
55
+
56
+ # scheduler settings
57
+ epoch = 800
58
+ optimizer = dict(type="AdamW", lr=0.006, weight_decay=0.05)
59
+ scheduler = dict(
60
+ type="OneCycleLR",
61
+ max_lr=[0.006, 0.0006],
62
+ pct_start=0.05,
63
+ anneal_strategy="cos",
64
+ div_factor=10.0,
65
+ final_div_factor=1000.0,
66
+ )
67
+ param_dicts = [dict(keyword="block", lr=0.0006)]
68
+
69
+ # dataset settings
70
+ dataset_type = "ARKitScenesLabelMakerConsensusDataset"
71
+ data_root = "data/alc"
72
+
73
+ data = dict(
74
+ num_classes=185,
75
+ ignore_index=-1,
76
+ names=WORDNET_NAMES,
77
+ train=dict(
78
+ type=dataset_type,
79
+ split="train",
80
+ data_root=data_root,
81
+ transform=[
82
+ dict(type="CenterShift", apply_z=True),
83
+ dict(type="RandomDropout", dropout_ratio=0.2, dropout_application_ratio=0.2),
84
+ # dict(type="RandomRotateTargetAngle", angle=(1/2, 1, 3/2), center=[0, 0, 0], axis="z", p=0.75),
85
+ dict(type="RandomRotate", angle=[-1, 1], axis="z", center=[0, 0, 0], p=0.5),
86
+ dict(type="RandomRotate", angle=[-1 / 64, 1 / 64], axis="x", p=0.5),
87
+ dict(type="RandomRotate", angle=[-1 / 64, 1 / 64], axis="y", p=0.5),
88
+ dict(type="RandomScale", scale=[0.9, 1.1]),
89
+ # dict(type="RandomShift", shift=[0.2, 0.2, 0.2]),
90
+ dict(type="RandomFlip", p=0.5),
91
+ dict(type="RandomJitter", sigma=0.005, clip=0.02),
92
+ dict(type="ElasticDistortion", distortion_params=[[0.2, 0.4], [0.8, 1.6]]),
93
+ dict(type="ChromaticAutoContrast", p=0.2, blend_factor=None),
94
+ dict(type="ChromaticTranslation", p=0.95, ratio=0.05),
95
+ dict(type="ChromaticJitter", p=0.95, std=0.05),
96
+ # dict(type="HueSaturationTranslation", hue_max=0.2, saturation_max=0.2),
97
+ # dict(type="RandomColorDrop", p=0.2, color_augment=0.0),
98
+ dict(
99
+ type="GridSample",
100
+ grid_size=0.02,
101
+ hash_type="fnv",
102
+ mode="train",
103
+ return_grid_coord=True,
104
+ ),
105
+ dict(type="SphereCrop", point_max=102400, mode="random"),
106
+ dict(type="CenterShift", apply_z=False),
107
+ dict(type="NormalizeColor"),
108
+ # dict(type="ShufflePoint"),
109
+ dict(type="ToTensor"),
110
+ dict(
111
+ type="Collect",
112
+ keys=("coord", "grid_coord", "segment"),
113
+ feat_keys=("color", "normal"),
114
+ ),
115
+ ],
116
+ test_mode=False,
117
+ ),
118
+ val=dict(
119
+ type=dataset_type,
120
+ split="val",
121
+ data_root=data_root,
122
+ transform=[
123
+ dict(type="CenterShift", apply_z=True),
124
+ dict(
125
+ type="GridSample",
126
+ grid_size=0.02,
127
+ hash_type="fnv",
128
+ mode="train",
129
+ return_grid_coord=True,
130
+ ),
131
+ dict(type="CenterShift", apply_z=False),
132
+ dict(type="NormalizeColor"),
133
+ dict(type="ToTensor"),
134
+ dict(
135
+ type="Collect",
136
+ keys=("coord", "grid_coord", "segment"),
137
+ feat_keys=("color", "normal"),
138
+ ),
139
+ ],
140
+ test_mode=False,
141
+ ),
142
+ test=dict(
143
+ type=dataset_type,
144
+ split="val",
145
+ data_root=data_root,
146
+ transform=[
147
+ dict(type="CenterShift", apply_z=True),
148
+ dict(type="NormalizeColor"),
149
+ ],
150
+ test_mode=True,
151
+ test_cfg=dict(
152
+ voxelize=dict(
153
+ type="GridSample",
154
+ grid_size=0.02,
155
+ hash_type="fnv",
156
+ mode="test",
157
+ keys=("coord", "color", "normal"),
158
+ return_grid_coord=True,
159
+ ),
160
+ crop=None,
161
+ post_transform=[
162
+ dict(type="CenterShift", apply_z=False),
163
+ dict(type="ToTensor"),
164
+ dict(
165
+ type="Collect",
166
+ keys=("coord", "grid_coord", "index"),
167
+ feat_keys=("color", "normal"),
168
+ ),
169
+ ],
170
+ aug_transform=[
171
+ [
172
+ dict(
173
+ type="RandomRotateTargetAngle",
174
+ angle=[0],
175
+ axis="z",
176
+ center=[0, 0, 0],
177
+ p=1,
178
+ )
179
+ ],
180
+ [
181
+ dict(
182
+ type="RandomRotateTargetAngle",
183
+ angle=[1 / 2],
184
+ axis="z",
185
+ center=[0, 0, 0],
186
+ p=1,
187
+ )
188
+ ],
189
+ [
190
+ dict(
191
+ type="RandomRotateTargetAngle",
192
+ angle=[1],
193
+ axis="z",
194
+ center=[0, 0, 0],
195
+ p=1,
196
+ )
197
+ ],
198
+ [
199
+ dict(
200
+ type="RandomRotateTargetAngle",
201
+ angle=[3 / 2],
202
+ axis="z",
203
+ center=[0, 0, 0],
204
+ p=1,
205
+ )
206
+ ],
207
+ [
208
+ dict(
209
+ type="RandomRotateTargetAngle",
210
+ angle=[0],
211
+ axis="z",
212
+ center=[0, 0, 0],
213
+ p=1,
214
+ ),
215
+ dict(type="RandomScale", scale=[0.95, 0.95]),
216
+ ],
217
+ [
218
+ dict(
219
+ type="RandomRotateTargetAngle",
220
+ angle=[1 / 2],
221
+ axis="z",
222
+ center=[0, 0, 0],
223
+ p=1,
224
+ ),
225
+ dict(type="RandomScale", scale=[0.95, 0.95]),
226
+ ],
227
+ [
228
+ dict(
229
+ type="RandomRotateTargetAngle",
230
+ angle=[1],
231
+ axis="z",
232
+ center=[0, 0, 0],
233
+ p=1,
234
+ ),
235
+ dict(type="RandomScale", scale=[0.95, 0.95]),
236
+ ],
237
+ [
238
+ dict(
239
+ type="RandomRotateTargetAngle",
240
+ angle=[3 / 2],
241
+ axis="z",
242
+ center=[0, 0, 0],
243
+ p=1,
244
+ ),
245
+ dict(type="RandomScale", scale=[0.95, 0.95]),
246
+ ],
247
+ [
248
+ dict(
249
+ type="RandomRotateTargetAngle",
250
+ angle=[0],
251
+ axis="z",
252
+ center=[0, 0, 0],
253
+ p=1,
254
+ ),
255
+ dict(type="RandomScale", scale=[1.05, 1.05]),
256
+ ],
257
+ [
258
+ dict(
259
+ type="RandomRotateTargetAngle",
260
+ angle=[1 / 2],
261
+ axis="z",
262
+ center=[0, 0, 0],
263
+ p=1,
264
+ ),
265
+ dict(type="RandomScale", scale=[1.05, 1.05]),
266
+ ],
267
+ [
268
+ dict(
269
+ type="RandomRotateTargetAngle",
270
+ angle=[1],
271
+ axis="z",
272
+ center=[0, 0, 0],
273
+ p=1,
274
+ ),
275
+ dict(type="RandomScale", scale=[1.05, 1.05]),
276
+ ],
277
+ [
278
+ dict(
279
+ type="RandomRotateTargetAngle",
280
+ angle=[3 / 2],
281
+ axis="z",
282
+ center=[0, 0, 0],
283
+ p=1,
284
+ ),
285
+ dict(type="RandomScale", scale=[1.05, 1.05]),
286
+ ],
287
+ [dict(type="RandomFlip", p=1)],
288
+ ],
289
+ ),
290
+ ),
291
+ )
configs/alc/semseg-pt-v3m1-0-base-wn199.py ADDED
@@ -0,0 +1,291 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from pointcept.datasets.preprocessing.alc.preprocess_arkitscenes_labelmaker_consensus import WORDNET_NAMES
2
+
3
+ _base_ = ["../_base_/default_runtime.py"]
4
+
5
+ # misc custom setting
6
+ batch_size = 12 # bs: total bs in all gpus
7
+ num_worker = 24
8
+ mix_prob = 0.8
9
+ empty_cache = False
10
+ enable_amp = True
11
+
12
+ # model settings
13
+ model = dict(
14
+ type="DefaultSegmentorV2",
15
+ num_classes=185,
16
+ backbone_out_channels=64,
17
+ backbone=dict(
18
+ type="PT-v3m1",
19
+ in_channels=6,
20
+ order=["z", "z-trans", "hilbert", "hilbert-trans"],
21
+ stride=(2, 2, 2, 2),
22
+ enc_depths=(2, 2, 2, 6, 2),
23
+ enc_channels=(32, 64, 128, 256, 512),
24
+ enc_num_head=(2, 4, 8, 16, 32),
25
+ enc_patch_size=(1024, 1024, 1024, 1024, 1024),
26
+ dec_depths=(2, 2, 2, 2),
27
+ dec_channels=(64, 64, 128, 256),
28
+ dec_num_head=(4, 4, 8, 16),
29
+ dec_patch_size=(1024, 1024, 1024, 1024),
30
+ mlp_ratio=4,
31
+ qkv_bias=True,
32
+ qk_scale=None,
33
+ attn_drop=0.0,
34
+ proj_drop=0.0,
35
+ drop_path=0.3,
36
+ shuffle_orders=True,
37
+ pre_norm=True,
38
+ enable_rpe=False,
39
+ enable_flash=True,
40
+ upcast_attention=False,
41
+ upcast_softmax=False,
42
+ cls_mode=False,
43
+ pdnorm_bn=False,
44
+ pdnorm_ln=False,
45
+ pdnorm_decouple=True,
46
+ pdnorm_adaptive=False,
47
+ pdnorm_affine=True,
48
+ pdnorm_conditions=("ScanNet", "S3DIS", "Structured3D"),
49
+ ),
50
+ criteria=[
51
+ dict(type="CrossEntropyLoss", loss_weight=1.0, ignore_index=-1),
52
+ dict(type="LovaszLoss", mode="multiclass", loss_weight=1.0, ignore_index=-1),
53
+ ],
54
+ )
55
+
56
+ # scheduler settings
57
+ epoch = 800
58
+ optimizer = dict(type="AdamW", lr=0.006, weight_decay=0.05)
59
+ scheduler = dict(
60
+ type="OneCycleLR",
61
+ max_lr=[0.006, 0.0006],
62
+ pct_start=0.05,
63
+ anneal_strategy="cos",
64
+ div_factor=10.0,
65
+ final_div_factor=1000.0,
66
+ )
67
+ param_dicts = [dict(keyword="block", lr=0.0006)]
68
+
69
+ # dataset settings
70
+ dataset_type = "ARKitScenesLabelMakerConsensusDataset"
71
+ data_root = "data/alc"
72
+
73
+ data = dict(
74
+ num_classes=185,
75
+ ignore_index=-1,
76
+ names=WORDNET_NAMES,
77
+ train=dict(
78
+ type=dataset_type,
79
+ split="train",
80
+ data_root=data_root,
81
+ transform=[
82
+ dict(type="CenterShift", apply_z=True),
83
+ dict(type="RandomDropout", dropout_ratio=0.2, dropout_application_ratio=0.2),
84
+ # dict(type="RandomRotateTargetAngle", angle=(1/2, 1, 3/2), center=[0, 0, 0], axis="z", p=0.75),
85
+ dict(type="RandomRotate", angle=[-1, 1], axis="z", center=[0, 0, 0], p=0.5),
86
+ dict(type="RandomRotate", angle=[-1 / 64, 1 / 64], axis="x", p=0.5),
87
+ dict(type="RandomRotate", angle=[-1 / 64, 1 / 64], axis="y", p=0.5),
88
+ dict(type="RandomScale", scale=[0.9, 1.1]),
89
+ # dict(type="RandomShift", shift=[0.2, 0.2, 0.2]),
90
+ dict(type="RandomFlip", p=0.5),
91
+ dict(type="RandomJitter", sigma=0.005, clip=0.02),
92
+ dict(type="ElasticDistortion", distortion_params=[[0.2, 0.4], [0.8, 1.6]]),
93
+ dict(type="ChromaticAutoContrast", p=0.2, blend_factor=None),
94
+ dict(type="ChromaticTranslation", p=0.95, ratio=0.05),
95
+ dict(type="ChromaticJitter", p=0.95, std=0.05),
96
+ # dict(type="HueSaturationTranslation", hue_max=0.2, saturation_max=0.2),
97
+ # dict(type="RandomColorDrop", p=0.2, color_augment=0.0),
98
+ dict(
99
+ type="GridSample",
100
+ grid_size=0.02,
101
+ hash_type="fnv",
102
+ mode="train",
103
+ return_grid_coord=True,
104
+ ),
105
+ dict(type="SphereCrop", point_max=102400, mode="random"),
106
+ dict(type="CenterShift", apply_z=False),
107
+ dict(type="NormalizeColor"),
108
+ # dict(type="ShufflePoint"),
109
+ dict(type="ToTensor"),
110
+ dict(
111
+ type="Collect",
112
+ keys=("coord", "grid_coord", "segment"),
113
+ feat_keys=("color", "normal"),
114
+ ),
115
+ ],
116
+ test_mode=False,
117
+ ),
118
+ val=dict(
119
+ type=dataset_type,
120
+ split="val",
121
+ data_root=data_root,
122
+ transform=[
123
+ dict(type="CenterShift", apply_z=True),
124
+ dict(
125
+ type="GridSample",
126
+ grid_size=0.02,
127
+ hash_type="fnv",
128
+ mode="train",
129
+ return_grid_coord=True,
130
+ ),
131
+ dict(type="CenterShift", apply_z=False),
132
+ dict(type="NormalizeColor"),
133
+ dict(type="ToTensor"),
134
+ dict(
135
+ type="Collect",
136
+ keys=("coord", "grid_coord", "segment"),
137
+ feat_keys=("color", "normal"),
138
+ ),
139
+ ],
140
+ test_mode=False,
141
+ ),
142
+ test=dict(
143
+ type=dataset_type,
144
+ split="val",
145
+ data_root=data_root,
146
+ transform=[
147
+ dict(type="CenterShift", apply_z=True),
148
+ dict(type="NormalizeColor"),
149
+ ],
150
+ test_mode=True,
151
+ test_cfg=dict(
152
+ voxelize=dict(
153
+ type="GridSample",
154
+ grid_size=0.02,
155
+ hash_type="fnv",
156
+ mode="test",
157
+ keys=("coord", "color", "normal"),
158
+ return_grid_coord=True,
159
+ ),
160
+ crop=None,
161
+ post_transform=[
162
+ dict(type="CenterShift", apply_z=False),
163
+ dict(type="ToTensor"),
164
+ dict(
165
+ type="Collect",
166
+ keys=("coord", "grid_coord", "index"),
167
+ feat_keys=("color", "normal"),
168
+ ),
169
+ ],
170
+ aug_transform=[
171
+ [
172
+ dict(
173
+ type="RandomRotateTargetAngle",
174
+ angle=[0],
175
+ axis="z",
176
+ center=[0, 0, 0],
177
+ p=1,
178
+ )
179
+ ],
180
+ [
181
+ dict(
182
+ type="RandomRotateTargetAngle",
183
+ angle=[1 / 2],
184
+ axis="z",
185
+ center=[0, 0, 0],
186
+ p=1,
187
+ )
188
+ ],
189
+ [
190
+ dict(
191
+ type="RandomRotateTargetAngle",
192
+ angle=[1],
193
+ axis="z",
194
+ center=[0, 0, 0],
195
+ p=1,
196
+ )
197
+ ],
198
+ [
199
+ dict(
200
+ type="RandomRotateTargetAngle",
201
+ angle=[3 / 2],
202
+ axis="z",
203
+ center=[0, 0, 0],
204
+ p=1,
205
+ )
206
+ ],
207
+ [
208
+ dict(
209
+ type="RandomRotateTargetAngle",
210
+ angle=[0],
211
+ axis="z",
212
+ center=[0, 0, 0],
213
+ p=1,
214
+ ),
215
+ dict(type="RandomScale", scale=[0.95, 0.95]),
216
+ ],
217
+ [
218
+ dict(
219
+ type="RandomRotateTargetAngle",
220
+ angle=[1 / 2],
221
+ axis="z",
222
+ center=[0, 0, 0],
223
+ p=1,
224
+ ),
225
+ dict(type="RandomScale", scale=[0.95, 0.95]),
226
+ ],
227
+ [
228
+ dict(
229
+ type="RandomRotateTargetAngle",
230
+ angle=[1],
231
+ axis="z",
232
+ center=[0, 0, 0],
233
+ p=1,
234
+ ),
235
+ dict(type="RandomScale", scale=[0.95, 0.95]),
236
+ ],
237
+ [
238
+ dict(
239
+ type="RandomRotateTargetAngle",
240
+ angle=[3 / 2],
241
+ axis="z",
242
+ center=[0, 0, 0],
243
+ p=1,
244
+ ),
245
+ dict(type="RandomScale", scale=[0.95, 0.95]),
246
+ ],
247
+ [
248
+ dict(
249
+ type="RandomRotateTargetAngle",
250
+ angle=[0],
251
+ axis="z",
252
+ center=[0, 0, 0],
253
+ p=1,
254
+ ),
255
+ dict(type="RandomScale", scale=[1.05, 1.05]),
256
+ ],
257
+ [
258
+ dict(
259
+ type="RandomRotateTargetAngle",
260
+ angle=[1 / 2],
261
+ axis="z",
262
+ center=[0, 0, 0],
263
+ p=1,
264
+ ),
265
+ dict(type="RandomScale", scale=[1.05, 1.05]),
266
+ ],
267
+ [
268
+ dict(
269
+ type="RandomRotateTargetAngle",
270
+ angle=[1],
271
+ axis="z",
272
+ center=[0, 0, 0],
273
+ p=1,
274
+ ),
275
+ dict(type="RandomScale", scale=[1.05, 1.05]),
276
+ ],
277
+ [
278
+ dict(
279
+ type="RandomRotateTargetAngle",
280
+ angle=[3 / 2],
281
+ axis="z",
282
+ center=[0, 0, 0],
283
+ p=1,
284
+ ),
285
+ dict(type="RandomScale", scale=[1.05, 1.05]),
286
+ ],
287
+ [dict(type="RandomFlip", p=1)],
288
+ ],
289
+ ),
290
+ ),
291
+ )
configs/scannet/semseg-pt-v3m1-1-ppt-extreme-alc.py ADDED
@@ -0,0 +1,782 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _base_ = ["../_base_/default_runtime.py"]
2
+
3
+ # misc custom setting
4
+ batch_size = 24 # bs: total bs in all gpus
5
+ num_worker = 48
6
+ mix_prob = 0.8
7
+ empty_cache = False
8
+ enable_amp = True
9
+ find_unused_parameters = True
10
+
11
+ # trainer
12
+ train = dict(
13
+ type="MultiDatasetTrainer",
14
+ )
15
+
16
+ # model
17
+ model = dict(
18
+ type="PPT-v1m1",
19
+ backbone=dict(
20
+ type="PT-v3m1",
21
+ in_channels=6,
22
+ order=("z", "z-trans", "hilbert", "hilbert-trans"),
23
+ stride=(2, 2, 2, 2),
24
+ enc_depths=(3, 3, 3, 6, 3),
25
+ enc_channels=(48, 96, 192, 384, 512),
26
+ enc_num_head=(3, 6, 12, 24, 32),
27
+ enc_patch_size=(1024, 1024, 1024, 1024, 1024),
28
+ dec_depths=(3, 3, 3, 3),
29
+ dec_channels=(64, 96, 192, 384),
30
+ dec_num_head=(4, 6, 12, 24),
31
+ dec_patch_size=(1024, 1024, 1024, 1024),
32
+ mlp_ratio=4,
33
+ qkv_bias=True,
34
+ qk_scale=None,
35
+ attn_drop=0.0,
36
+ proj_drop=0.0,
37
+ drop_path=0.3,
38
+ shuffle_orders=True,
39
+ pre_norm=True,
40
+ enable_rpe=False,
41
+ enable_flash=True,
42
+ upcast_attention=False,
43
+ upcast_softmax=False,
44
+ cls_mode=False,
45
+ pdnorm_bn=True,
46
+ pdnorm_ln=True,
47
+ pdnorm_decouple=True,
48
+ pdnorm_adaptive=False,
49
+ pdnorm_affine=True,
50
+ pdnorm_conditions=(
51
+ "S3DIS",
52
+ "ScanNet",
53
+ "Structured3D",
54
+ "ALC",
55
+ # "ScanNet200"
56
+ ),
57
+ ),
58
+ criteria=[dict(type="CrossEntropyLoss", loss_weight=1.0, ignore_index=-1), dict(type="LovaszLoss", mode="multiclass", loss_weight=1.0, ignore_index=-1)],
59
+ backbone_out_channels=64,
60
+ context_channels=256,
61
+ conditions=(
62
+ "S3DIS",
63
+ "ScanNet",
64
+ "Structured3D",
65
+ "ALC",
66
+ # "ScanNet200"
67
+ ),
68
+ template="[x]",
69
+ clip_model="ViT-B/16",
70
+ class_name=(
71
+ "wall",
72
+ "floor",
73
+ "cabinet",
74
+ "bed",
75
+ "chair",
76
+ "sofa",
77
+ "table",
78
+ "door",
79
+ "window",
80
+ "bookshelf",
81
+ "bookcase",
82
+ "picture",
83
+ "counter",
84
+ "desk",
85
+ "shelves",
86
+ "curtain",
87
+ "dresser",
88
+ "pillow",
89
+ "mirror",
90
+ "ceiling",
91
+ "refrigerator",
92
+ "television",
93
+ "shower curtain",
94
+ "nightstand",
95
+ "toilet",
96
+ "sink",
97
+ "lamp",
98
+ "bathtub",
99
+ "garbagebin",
100
+ "board",
101
+ "beam",
102
+ "column",
103
+ "clutter",
104
+ "otherstructure",
105
+ "otherfurniture",
106
+ "otherprop",
107
+ "book",
108
+ "ashcan",
109
+ "display",
110
+ "cushion",
111
+ "box",
112
+ "doorframe",
113
+ "swivel chair",
114
+ "towel",
115
+ "backpack",
116
+ "chest of drawers",
117
+ "apparel",
118
+ "armchair",
119
+ "plant",
120
+ "radiator",
121
+ "toilet tissue",
122
+ "shoe",
123
+ "bag",
124
+ "bottle",
125
+ "countertop",
126
+ "coffee table",
127
+ "computer keyboard",
128
+ "fridge",
129
+ "stool",
130
+ "computer",
131
+ "mug",
132
+ "telephone",
133
+ "light",
134
+ "jacket",
135
+ "microwave",
136
+ "footstool",
137
+ "baggage",
138
+ "laptop",
139
+ "printer",
140
+ "shower stall",
141
+ "soap dispenser",
142
+ "stove",
143
+ "fan",
144
+ "paper",
145
+ "stand",
146
+ "bench",
147
+ "wardrobe",
148
+ "blanket",
149
+ "booth",
150
+ "duplicator",
151
+ "bar",
152
+ "soap dish",
153
+ "switch",
154
+ "coffee maker",
155
+ "decoration",
156
+ "range hood",
157
+ "blackboard",
158
+ "clock",
159
+ "railing",
160
+ "mat",
161
+ "seat",
162
+ "bannister",
163
+ "container",
164
+ "mouse",
165
+ "person",
166
+ "stairway",
167
+ "basket",
168
+ "dumbbell",
169
+ "bucket",
170
+ "windowsill",
171
+ "signboard",
172
+ "dishwasher",
173
+ "loudspeaker",
174
+ "washer",
175
+ "paper towel",
176
+ "clothes hamper",
177
+ "piano",
178
+ "sack",
179
+ "handcart",
180
+ "blind",
181
+ "dish rack",
182
+ "mailbox",
183
+ "bicycle",
184
+ "ladder",
185
+ "rack",
186
+ "tray",
187
+ "toaster",
188
+ "paper cutter",
189
+ "plunger",
190
+ "dryer",
191
+ "guitar",
192
+ "fire extinguisher",
193
+ "pitcher",
194
+ "pipe",
195
+ "plate",
196
+ "vacuum",
197
+ "bowl",
198
+ "hat",
199
+ "rod",
200
+ "water cooler",
201
+ "kettle",
202
+ "oven",
203
+ "scale",
204
+ "broom",
205
+ "hand blower",
206
+ "coatrack",
207
+ "teddy",
208
+ "alarm clock",
209
+ "ironing board",
210
+ "fire alarm",
211
+ "machine",
212
+ "music stand",
213
+ "fireplace",
214
+ "furniture",
215
+ "vase",
216
+ "vent",
217
+ "candle",
218
+ "crate",
219
+ "dustpan",
220
+ "earphone",
221
+ "jar",
222
+ "projector",
223
+ "gat",
224
+ "step",
225
+ "step stool",
226
+ "vending machine",
227
+ "coat",
228
+ "coat hanger",
229
+ "drinking fountain",
230
+ "hamper",
231
+ "thermostat",
232
+ "banner",
233
+ "iron",
234
+ "soap",
235
+ "chopping board",
236
+ "kitchen island",
237
+ "shirt",
238
+ "sleeping bag",
239
+ "tire",
240
+ "toothbrush",
241
+ "bathrobe",
242
+ "faucet",
243
+ "slipper",
244
+ "thermos",
245
+ "tripod",
246
+ "dispenser",
247
+ "heater",
248
+ "pool table",
249
+ "remote control",
250
+ "stapler",
251
+ "treadmill",
252
+ "beanbag",
253
+ "dartboard",
254
+ "metronome",
255
+ "rope",
256
+ "sewing machine",
257
+ "shredder",
258
+ "toolbox",
259
+ "water heater",
260
+ "brush",
261
+ "control",
262
+ "dais",
263
+ "dollhouse",
264
+ "envelope",
265
+ "food",
266
+ "frying pan",
267
+ "helmet",
268
+ "tennis racket",
269
+ "umbrella",
270
+ "couch",
271
+ "shelf",
272
+ "office chair",
273
+ "monitor",
274
+ "kitchen cabinet",
275
+ "clothes",
276
+ "tv",
277
+ "end table",
278
+ "dining table",
279
+ "keyboard",
280
+ "toilet paper",
281
+ "tv stand",
282
+ "whiteboard",
283
+ "trash can",
284
+ "closet",
285
+ "stairs",
286
+ "computer tower",
287
+ "bin",
288
+ "ottoman",
289
+ "washing machine",
290
+ "copier",
291
+ "sofa chair",
292
+ "file cabinet",
293
+ "shower",
294
+ "paper towel dispenser",
295
+ "blinds",
296
+ "suitcase",
297
+ "rail",
298
+ "recycling bin",
299
+ "laundry basket",
300
+ "clothes dryer",
301
+ "toilet paper holder",
302
+ "speaker",
303
+ "bathroom stall",
304
+ "shower wall",
305
+ "cup",
306
+ "storage bin",
307
+ "paper towel roll",
308
+ "bulletin board",
309
+ "kitchen counter",
310
+ "toilet paper dispenser",
311
+ "mini fridge",
312
+ "ball",
313
+ "shower curtain rod",
314
+ "shower door",
315
+ "pillar",
316
+ "ledge",
317
+ "toaster oven",
318
+ "toilet seat cover dispenser",
319
+ "cart",
320
+ "storage container",
321
+ "tissue box",
322
+ "light switch",
323
+ "power outlet",
324
+ "sign",
325
+ "closet door",
326
+ "vacuum cleaner",
327
+ "stuffed animal",
328
+ "headphones",
329
+ "guitar case",
330
+ "hair dryer",
331
+ "water bottle",
332
+ "handicap bar",
333
+ "purse",
334
+ "shower floor",
335
+ "water pitcher",
336
+ "paper bag",
337
+ "projector screen",
338
+ "divider",
339
+ "laundry detergent",
340
+ "bathroom counter",
341
+ "object",
342
+ "bathroom vanity",
343
+ "closet wall",
344
+ "laundry hamper",
345
+ "bathroom stall door",
346
+ "ceiling light",
347
+ "trash bin",
348
+ "stair rail",
349
+ "tube",
350
+ "bathroom cabinet",
351
+ "cd case",
352
+ "closet rod",
353
+ "coffee kettle",
354
+ "structure",
355
+ "shower head",
356
+ "keyboard piano",
357
+ "case of water bottles",
358
+ "coat rack",
359
+ "storage organizer",
360
+ "folded chair",
361
+ "power strip",
362
+ "calendar",
363
+ "poster",
364
+ "potted plant",
365
+ "luggage",
366
+ "mattress",
367
+ ),
368
+ valid_index=(
369
+ (0, 1, 4, 5, 6, 7, 8, 10, 19, 29, 30, 31, 32),
370
+ (0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 11, 12, 13, 15, 20, 22, 24, 25, 27, 34),
371
+ (0, 1, 2, 3, 4, 5, 6, 7, 8, 11, 13, 14, 15, 16, 17, 18, 19, 20, 21, 23, 25, 26, 33, 34, 35),
372
+ (
373
+ 0,
374
+ 4,
375
+ 36,
376
+ 2,
377
+ 7,
378
+ 1,
379
+ 37,
380
+ 6,
381
+ 8,
382
+ 9,
383
+ 38,
384
+ 39,
385
+ 40,
386
+ 11,
387
+ 19,
388
+ 41,
389
+ 13,
390
+ 42,
391
+ 43,
392
+ 5,
393
+ 25,
394
+ 44,
395
+ 26,
396
+ 45,
397
+ 46,
398
+ 47,
399
+ 3,
400
+ 15,
401
+ 18,
402
+ 48,
403
+ 49,
404
+ 50,
405
+ 51,
406
+ 52,
407
+ 53,
408
+ 54,
409
+ 55,
410
+ 24,
411
+ 56,
412
+ 57,
413
+ 58,
414
+ 59,
415
+ 60,
416
+ 61,
417
+ 62,
418
+ 63,
419
+ 27,
420
+ 22,
421
+ 64,
422
+ 65,
423
+ 66,
424
+ 67,
425
+ 68,
426
+ 69,
427
+ 70,
428
+ 71,
429
+ 72,
430
+ 73,
431
+ 74,
432
+ 75,
433
+ 76,
434
+ 77,
435
+ 78,
436
+ 79,
437
+ 80,
438
+ 81,
439
+ 82,
440
+ 83,
441
+ 84,
442
+ 85,
443
+ 86,
444
+ 87,
445
+ 88,
446
+ 89,
447
+ 90,
448
+ 91,
449
+ 92,
450
+ 93,
451
+ 94,
452
+ 95,
453
+ 96,
454
+ 97,
455
+ 31,
456
+ 98,
457
+ 99,
458
+ 100,
459
+ 101,
460
+ 102,
461
+ 103,
462
+ 104,
463
+ 105,
464
+ 106,
465
+ 107,
466
+ 108,
467
+ 109,
468
+ 110,
469
+ 111,
470
+ 52,
471
+ 112,
472
+ 113,
473
+ 114,
474
+ 115,
475
+ 116,
476
+ 117,
477
+ 118,
478
+ 119,
479
+ 120,
480
+ 121,
481
+ 122,
482
+ 123,
483
+ 124,
484
+ 125,
485
+ 126,
486
+ 127,
487
+ 128,
488
+ 129,
489
+ 130,
490
+ 131,
491
+ 132,
492
+ 133,
493
+ 134,
494
+ 135,
495
+ 136,
496
+ 137,
497
+ 138,
498
+ 139,
499
+ 140,
500
+ 141,
501
+ 142,
502
+ 143,
503
+ 144,
504
+ 145,
505
+ 146,
506
+ 147,
507
+ 148,
508
+ 149,
509
+ 150,
510
+ 151,
511
+ 152,
512
+ 153,
513
+ 154,
514
+ 155,
515
+ 156,
516
+ 157,
517
+ 158,
518
+ 159,
519
+ 160,
520
+ 161,
521
+ 162,
522
+ 163,
523
+ 164,
524
+ 165,
525
+ 166,
526
+ 167,
527
+ 168,
528
+ 169,
529
+ 170,
530
+ 171,
531
+ 172,
532
+ 173,
533
+ 174,
534
+ 175,
535
+ 176,
536
+ 177,
537
+ 178,
538
+ 179,
539
+ 180,
540
+ 181,
541
+ 182,
542
+ 183,
543
+ 184,
544
+ 185,
545
+ 186,
546
+ 187,
547
+ 188,
548
+ 189,
549
+ 190,
550
+ 191,
551
+ 192,
552
+ 193,
553
+ 194,
554
+ 195,
555
+ 196,
556
+ 197,
557
+ 198,
558
+ ),
559
+ ),
560
+ backbone_mode=False,
561
+ )
562
+
563
+ # optimizer
564
+ # epoch = 800
565
+ # eval_epoch = 800
566
+ epoch = 1000
567
+ eval_epoch = 1000
568
+ # epoch = 1600
569
+ # eval_epoch = 1600
570
+ optimizer = dict(type="AdamW", lr=0.005, weight_decay=0.05)
571
+ scheduler = dict(
572
+ type="OneCycleLR",
573
+ max_lr=[0.005, 0.0005],
574
+ pct_start=0.05,
575
+ anneal_strategy="cos",
576
+ div_factor=10.0,
577
+ final_div_factor=1000.0,
578
+ )
579
+ param_dicts = [dict(keyword="block", lr=0.0005)]
580
+
581
+ # datasets
582
+ data = dict(
583
+ num_classes=20,
584
+ ignore_index=-1,
585
+ names=["wall", "floor", "cabinet", "bed", "chair", "sofa", "table", "door", "window", "bookshelf", "picture", "counter", "desk", "curtain", "refridgerator", "shower curtain", "toilet", "sink", "bathtub", "otherfurniture"],
586
+ train=dict(
587
+ type="ConcatDataset",
588
+ datasets=[
589
+ # # Structured3DDataset
590
+ # dict(
591
+ # type="Structured3DDataset",
592
+ # split=["train", "val", "test"],
593
+ # data_root="data/structured3d",
594
+ # transform=[
595
+ # dict(type="CenterShift", apply_z=True),
596
+ # dict(type="RandomDropout", dropout_ratio=0.2, dropout_application_ratio=0.2),
597
+ # dict(type="RandomRotate", angle=[-1, 1], axis="z", center=[0, 0, 0], p=0.5),
598
+ # dict(type="RandomRotate", angle=[-0.015625, 0.015625], axis="x", p=0.5),
599
+ # dict(type="RandomRotate", angle=[-0.015625, 0.015625], axis="y", p=0.5),
600
+ # dict(type="RandomScale", scale=[0.9, 1.1]),
601
+ # dict(type="RandomFlip", p=0.5),
602
+ # dict(type="RandomJitter", sigma=0.005, clip=0.02),
603
+ # dict(type="ElasticDistortion", distortion_params=[[0.2, 0.4], [0.8, 1.6]]),
604
+ # dict(type="ChromaticAutoContrast", p=0.2, blend_factor=None),
605
+ # dict(type="ChromaticTranslation", p=0.95, ratio=0.05),
606
+ # dict(type="ChromaticJitter", p=0.95, std=0.05),
607
+ # dict(type="GridSample", grid_size=0.02, hash_type="fnv", mode="train", return_grid_coord=True),
608
+ # dict(type="SphereCrop", sample_rate=0.8, mode="random"),
609
+ # dict(type="SphereCrop", point_max=102400, mode="random"),
610
+ # dict(type="CenterShift", apply_z=False),
611
+ # dict(type="NormalizeColor"),
612
+ # dict(type="Add", keys_dict=dict(condition="Structured3D")),
613
+ # dict(type="ToTensor"),
614
+ # dict(type="Collect", keys=("coord", "grid_coord", "segment", "condition"), feat_keys=("color", "normal")),
615
+ # ],
616
+ # test_mode=False,
617
+ # loop=1,
618
+ # ),
619
+ # ScanNetDataset
620
+ dict(
621
+ type="ScanNetDataset",
622
+ split="train",
623
+ data_root="data/scannet",
624
+ transform=[
625
+ dict(type="CenterShift", apply_z=True),
626
+ dict(type="RandomDropout", dropout_ratio=0.2, dropout_application_ratio=0.2),
627
+ dict(type="RandomRotate", angle=[-1, 1], axis="z", center=[0, 0, 0], p=0.5),
628
+ dict(type="RandomRotate", angle=[-0.015625, 0.015625], axis="x", p=0.5),
629
+ dict(type="RandomRotate", angle=[-0.015625, 0.015625], axis="y", p=0.5),
630
+ dict(type="RandomScale", scale=[0.9, 1.1]),
631
+ dict(type="RandomFlip", p=0.5),
632
+ dict(type="RandomJitter", sigma=0.005, clip=0.02),
633
+ dict(type="ElasticDistortion", distortion_params=[[0.2, 0.4], [0.8, 1.6]]),
634
+ dict(type="ChromaticAutoContrast", p=0.2, blend_factor=None),
635
+ dict(type="ChromaticTranslation", p=0.95, ratio=0.05),
636
+ dict(type="ChromaticJitter", p=0.95, std=0.05),
637
+ dict(type="GridSample", grid_size=0.02, hash_type="fnv", mode="train", return_grid_coord=True),
638
+ dict(type="SphereCrop", point_max=102400, mode="random"),
639
+ dict(type="CenterShift", apply_z=False),
640
+ dict(type="NormalizeColor"),
641
+ dict(type="ShufflePoint"),
642
+ dict(type="Add", keys_dict=dict(condition="ScanNet")),
643
+ dict(type="ToTensor"),
644
+ dict(type="Collect", keys=("coord", "grid_coord", "segment", "condition"), feat_keys=("color", "normal")),
645
+ ],
646
+ test_mode=False,
647
+ loop=1,
648
+ ),
649
+ # S3DISDataset
650
+ dict(
651
+ type="S3DISDataset",
652
+ split=("Area_1", "Area_2", "Area_3", "Area_4", "Area_6"),
653
+ data_root="data/s3dis",
654
+ transform=[
655
+ dict(type="CenterShift", apply_z=True),
656
+ dict(type="RandomDropout", dropout_ratio=0.2, dropout_application_ratio=0.2),
657
+ dict(type="RandomRotate", angle=[-1, 1], axis="z", center=[0, 0, 0], p=0.5),
658
+ dict(type="RandomRotate", angle=[-0.015625, 0.015625], axis="x", p=0.5),
659
+ dict(type="RandomRotate", angle=[-0.015625, 0.015625], axis="y", p=0.5),
660
+ dict(type="RandomScale", scale=[0.9, 1.1]),
661
+ dict(type="RandomFlip", p=0.5),
662
+ dict(type="RandomJitter", sigma=0.005, clip=0.02),
663
+ dict(type="ChromaticAutoContrast", p=0.2, blend_factor=None),
664
+ dict(type="ChromaticTranslation", p=0.95, ratio=0.05),
665
+ dict(type="ChromaticJitter", p=0.95, std=0.05),
666
+ dict(type="GridSample", grid_size=0.02, hash_type="fnv", mode="train", return_grid_coord=True),
667
+ dict(type="SphereCrop", sample_rate=0.6, mode="random"),
668
+ dict(type="SphereCrop", point_max=204800, mode="random"),
669
+ dict(type="CenterShift", apply_z=False),
670
+ dict(type="NormalizeColor"),
671
+ dict(type="Add", keys_dict=dict(condition="S3DIS")),
672
+ dict(type="ToTensor"),
673
+ dict(type="Collect", keys=("coord", "grid_coord", "segment", "condition"), feat_keys=("color", "normal")),
674
+ ],
675
+ test_mode=False,
676
+ loop=1,
677
+ ),
678
+ # ALC
679
+ dict(
680
+ type="ARKitScenesLabelMakerConsensusDataset",
681
+ split=["train", "val"],
682
+ data_root="data/alc",
683
+ transform=[
684
+ dict(type="CenterShift", apply_z=True),
685
+ dict(type="RandomDropout", dropout_ratio=0.2, dropout_application_ratio=0.2),
686
+ # dict(type="RandomRotateTargetAngle", angle=(1/2, 1, 3/2), center=[0, 0, 0], axis="z", p=0.75),
687
+ dict(type="RandomRotate", angle=[-1, 1], axis="z", center=[0, 0, 0], p=0.5),
688
+ dict(type="RandomRotate", angle=[-1 / 64, 1 / 64], axis="x", p=0.5),
689
+ dict(type="RandomRotate", angle=[-1 / 64, 1 / 64], axis="y", p=0.5),
690
+ dict(type="RandomScale", scale=[0.9, 1.1]),
691
+ # dict(type="RandomShift", shift=[0.2, 0.2, 0.2]),
692
+ dict(type="RandomFlip", p=0.5),
693
+ dict(type="RandomJitter", sigma=0.005, clip=0.02),
694
+ dict(type="ElasticDistortion", distortion_params=[[0.2, 0.4], [0.8, 1.6]]),
695
+ dict(type="ChromaticAutoContrast", p=0.2, blend_factor=None),
696
+ dict(type="ChromaticTranslation", p=0.95, ratio=0.05),
697
+ dict(type="ChromaticJitter", p=0.95, std=0.05),
698
+ # dict(type="HueSaturationTranslation", hue_max=0.2, saturation_max=0.2),
699
+ # dict(type="RandomColorDrop", p=0.2, color_augment=0.0),
700
+ dict(
701
+ type="GridSample",
702
+ grid_size=0.02,
703
+ hash_type="fnv",
704
+ mode="train",
705
+ return_grid_coord=True,
706
+ ),
707
+ dict(type="SphereCrop", point_max=102400, mode="random"),
708
+ dict(type="CenterShift", apply_z=False),
709
+ dict(type="NormalizeColor"),
710
+ # dict(type="ShufflePoint"),
711
+ dict(type="Add", keys_dict=dict(condition="ALC")),
712
+ dict(type="ToTensor"),
713
+ dict(
714
+ type="Collect",
715
+ keys=("coord", "grid_coord", "segment", "condition"),
716
+ feat_keys=("color", "normal"),
717
+ ),
718
+ ],
719
+ test_mode=False,
720
+ loop=2,
721
+ ),
722
+ ],
723
+ loop=1,
724
+ ),
725
+ val=dict(
726
+ type="ScanNetDataset",
727
+ split="val",
728
+ data_root="data/scannet",
729
+ transform=[
730
+ dict(type="CenterShift", apply_z=True),
731
+ dict(type="GridSample", grid_size=0.02, hash_type="fnv", mode="train", return_grid_coord=True),
732
+ dict(type="CenterShift", apply_z=False),
733
+ dict(type="NormalizeColor"),
734
+ dict(type="ToTensor"),
735
+ dict(type="Add", keys_dict=dict(condition="ScanNet")),
736
+ dict(type="Collect", keys=("coord", "grid_coord", "segment", "condition"), feat_keys=("color", "normal")),
737
+ ],
738
+ test_mode=False,
739
+ ),
740
+ test=dict(
741
+ type="ScanNetDataset",
742
+ split="val",
743
+ data_root="data/scannet",
744
+ transform=[dict(type="CenterShift", apply_z=True), dict(type="NormalizeColor")],
745
+ test_mode=True,
746
+ test_cfg=dict(
747
+ voxelize=dict(type="GridSample", grid_size=0.02, hash_type="fnv", mode="test", keys=("coord", "color", "normal"), return_grid_coord=True),
748
+ crop=None,
749
+ post_transform=[
750
+ dict(type="CenterShift", apply_z=False),
751
+ dict(type="Add", keys_dict=dict(condition="ScanNet")),
752
+ dict(type="ToTensor"),
753
+ dict(type="Collect", keys=("coord", "grid_coord", "index", "condition"), feat_keys=("color", "normal")),
754
+ ],
755
+ aug_transform=[
756
+ [{"type": "RandomRotateTargetAngle", "angle": [0], "axis": "z", "center": [0, 0, 0], "p": 1}],
757
+ [{"type": "RandomRotateTargetAngle", "angle": [0.5], "axis": "z", "center": [0, 0, 0], "p": 1}],
758
+ [{"type": "RandomRotateTargetAngle", "angle": [1], "axis": "z", "center": [0, 0, 0], "p": 1}],
759
+ [{"type": "RandomRotateTargetAngle", "angle": [1.5], "axis": "z", "center": [0, 0, 0], "p": 1}],
760
+ [{"type": "RandomRotateTargetAngle", "angle": [0], "axis": "z", "center": [0, 0, 0], "p": 1}, {"type": "RandomScale", "scale": [0.95, 0.95]}],
761
+ [{"type": "RandomRotateTargetAngle", "angle": [0.5], "axis": "z", "center": [0, 0, 0], "p": 1}, {"type": "RandomScale", "scale": [0.95, 0.95]}],
762
+ [{"type": "RandomRotateTargetAngle", "angle": [1], "axis": "z", "center": [0, 0, 0], "p": 1}, {"type": "RandomScale", "scale": [0.95, 0.95]}],
763
+ [{"type": "RandomRotateTargetAngle", "angle": [1.5], "axis": "z", "center": [0, 0, 0], "p": 1}, {"type": "RandomScale", "scale": [0.95, 0.95]}],
764
+ [{"type": "RandomRotateTargetAngle", "angle": [0], "axis": "z", "center": [0, 0, 0], "p": 1}, {"type": "RandomScale", "scale": [1.05, 1.05]}],
765
+ [{"type": "RandomRotateTargetAngle", "angle": [0.5], "axis": "z", "center": [0, 0, 0], "p": 1}, {"type": "RandomScale", "scale": [1.05, 1.05]}],
766
+ [{"type": "RandomRotateTargetAngle", "angle": [1], "axis": "z", "center": [0, 0, 0], "p": 1}, {"type": "RandomScale", "scale": [1.05, 1.05]}],
767
+ [{"type": "RandomRotateTargetAngle", "angle": [1.5], "axis": "z", "center": [0, 0, 0], "p": 1}, {"type": "RandomScale", "scale": [1.05, 1.05]}],
768
+ [{"type": "RandomFlip", "p": 1}],
769
+ ],
770
+ ),
771
+ ),
772
+ )
773
+
774
+ # hook
775
+ hooks = [
776
+ dict(type="CheckpointLoader"),
777
+ dict(type="IterationTimer", warmup_iter=2),
778
+ dict(type="InformationWriter"),
779
+ dict(type="SemSegEvaluator"),
780
+ dict(type="CheckpointSaver", save_freq=None),
781
+ dict(type="PreciseEvaluator", test_last=True),
782
+ ]
configs/scannet200/semseg-pt-v3m1-1-ppt-extreme-alc.py ADDED
@@ -0,0 +1,972 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from pointcept.datasets.preprocessing.scannet.meta_data.scannet200_constants import CLASS_LABELS_200
2
+
3
+ _base_ = ["../_base_/default_runtime.py"]
4
+
5
+ # misc custom setting
6
+ batch_size = 24 # bs: total bs in all gpus
7
+ num_worker = 36
8
+ mix_prob = 0.8
9
+ empty_cache = False
10
+ enable_amp = True
11
+ find_unused_parameters = True
12
+
13
+ # trainer
14
+ train = dict(
15
+ type="MultiDatasetTrainer",
16
+ )
17
+
18
+ # model
19
+ model = dict(
20
+ type="PPT-v1m1",
21
+ backbone=dict(
22
+ type="PT-v3m1",
23
+ in_channels=6,
24
+ order=("z", "z-trans", "hilbert", "hilbert-trans"),
25
+ stride=(2, 2, 2, 2),
26
+ enc_depths=(3, 3, 3, 6, 3),
27
+ enc_channels=(48, 96, 192, 384, 512),
28
+ enc_num_head=(3, 6, 12, 24, 32),
29
+ enc_patch_size=(1024, 1024, 1024, 1024, 1024),
30
+ dec_depths=(3, 3, 3, 3),
31
+ dec_channels=(64, 96, 192, 384),
32
+ dec_num_head=(4, 6, 12, 24),
33
+ dec_patch_size=(1024, 1024, 1024, 1024),
34
+ mlp_ratio=4,
35
+ qkv_bias=True,
36
+ qk_scale=None,
37
+ attn_drop=0.0,
38
+ proj_drop=0.0,
39
+ drop_path=0.3,
40
+ shuffle_orders=True,
41
+ pre_norm=True,
42
+ enable_rpe=False,
43
+ enable_flash=True,
44
+ upcast_attention=False,
45
+ upcast_softmax=False,
46
+ cls_mode=False,
47
+ pdnorm_bn=True,
48
+ pdnorm_ln=True,
49
+ pdnorm_decouple=True,
50
+ pdnorm_adaptive=False,
51
+ pdnorm_affine=True,
52
+ pdnorm_conditions=(
53
+ "S3DIS",
54
+ # "ScanNet",
55
+ "Structured3D",
56
+ "ALC",
57
+ "ScanNet200",
58
+ ),
59
+ ),
60
+ criteria=[dict(type="CrossEntropyLoss", loss_weight=1.0, ignore_index=-1), dict(type="LovaszLoss", mode="multiclass", loss_weight=1.0, ignore_index=-1)],
61
+ backbone_out_channels=64,
62
+ context_channels=256,
63
+ conditions=(
64
+ "S3DIS",
65
+ # "ScanNet",
66
+ "Structured3D",
67
+ "ALC",
68
+ "ScanNet200",
69
+ ),
70
+ template="[x]",
71
+ clip_model="ViT-B/16",
72
+ class_name=(
73
+ "wall",
74
+ "floor",
75
+ "cabinet",
76
+ "bed",
77
+ "chair",
78
+ "sofa",
79
+ "table",
80
+ "door",
81
+ "window",
82
+ "bookshelf",
83
+ "bookcase",
84
+ "picture",
85
+ "counter",
86
+ "desk",
87
+ "shelves",
88
+ "curtain",
89
+ "dresser",
90
+ "pillow",
91
+ "mirror",
92
+ "ceiling",
93
+ "refrigerator",
94
+ "television",
95
+ "shower curtain",
96
+ "nightstand",
97
+ "toilet",
98
+ "sink",
99
+ "lamp",
100
+ "bathtub",
101
+ "garbagebin",
102
+ "board",
103
+ "beam",
104
+ "column",
105
+ "clutter",
106
+ "otherstructure",
107
+ "otherfurniture",
108
+ "otherprop",
109
+ "book",
110
+ "ashcan",
111
+ "display",
112
+ "cushion",
113
+ "box",
114
+ "doorframe",
115
+ "swivel chair",
116
+ "towel",
117
+ "backpack",
118
+ "chest of drawers",
119
+ "apparel",
120
+ "armchair",
121
+ "plant",
122
+ "radiator",
123
+ "toilet tissue",
124
+ "shoe",
125
+ "bag",
126
+ "bottle",
127
+ "countertop",
128
+ "coffee table",
129
+ "computer keyboard",
130
+ "fridge",
131
+ "stool",
132
+ "computer",
133
+ "mug",
134
+ "telephone",
135
+ "light",
136
+ "jacket",
137
+ "microwave",
138
+ "footstool",
139
+ "baggage",
140
+ "laptop",
141
+ "printer",
142
+ "shower stall",
143
+ "soap dispenser",
144
+ "stove",
145
+ "fan",
146
+ "paper",
147
+ "stand",
148
+ "bench",
149
+ "wardrobe",
150
+ "blanket",
151
+ "booth",
152
+ "duplicator",
153
+ "bar",
154
+ "soap dish",
155
+ "switch",
156
+ "coffee maker",
157
+ "decoration",
158
+ "range hood",
159
+ "blackboard",
160
+ "clock",
161
+ "railing",
162
+ "mat",
163
+ "seat",
164
+ "bannister",
165
+ "container",
166
+ "mouse",
167
+ "person",
168
+ "stairway",
169
+ "basket",
170
+ "dumbbell",
171
+ "bucket",
172
+ "windowsill",
173
+ "signboard",
174
+ "dishwasher",
175
+ "loudspeaker",
176
+ "washer",
177
+ "paper towel",
178
+ "clothes hamper",
179
+ "piano",
180
+ "sack",
181
+ "handcart",
182
+ "blind",
183
+ "dish rack",
184
+ "mailbox",
185
+ "bicycle",
186
+ "ladder",
187
+ "rack",
188
+ "tray",
189
+ "toaster",
190
+ "paper cutter",
191
+ "plunger",
192
+ "dryer",
193
+ "guitar",
194
+ "fire extinguisher",
195
+ "pitcher",
196
+ "pipe",
197
+ "plate",
198
+ "vacuum",
199
+ "bowl",
200
+ "hat",
201
+ "rod",
202
+ "water cooler",
203
+ "kettle",
204
+ "oven",
205
+ "scale",
206
+ "broom",
207
+ "hand blower",
208
+ "coatrack",
209
+ "teddy",
210
+ "alarm clock",
211
+ "ironing board",
212
+ "fire alarm",
213
+ "machine",
214
+ "music stand",
215
+ "fireplace",
216
+ "furniture",
217
+ "vase",
218
+ "vent",
219
+ "candle",
220
+ "crate",
221
+ "dustpan",
222
+ "earphone",
223
+ "jar",
224
+ "projector",
225
+ "gat",
226
+ "step",
227
+ "step stool",
228
+ "vending machine",
229
+ "coat",
230
+ "coat hanger",
231
+ "drinking fountain",
232
+ "hamper",
233
+ "thermostat",
234
+ "banner",
235
+ "iron",
236
+ "soap",
237
+ "chopping board",
238
+ "kitchen island",
239
+ "shirt",
240
+ "sleeping bag",
241
+ "tire",
242
+ "toothbrush",
243
+ "bathrobe",
244
+ "faucet",
245
+ "slipper",
246
+ "thermos",
247
+ "tripod",
248
+ "dispenser",
249
+ "heater",
250
+ "pool table",
251
+ "remote control",
252
+ "stapler",
253
+ "treadmill",
254
+ "beanbag",
255
+ "dartboard",
256
+ "metronome",
257
+ "rope",
258
+ "sewing machine",
259
+ "shredder",
260
+ "toolbox",
261
+ "water heater",
262
+ "brush",
263
+ "control",
264
+ "dais",
265
+ "dollhouse",
266
+ "envelope",
267
+ "food",
268
+ "frying pan",
269
+ "helmet",
270
+ "tennis racket",
271
+ "umbrella",
272
+ "couch",
273
+ "shelf",
274
+ "office chair",
275
+ "monitor",
276
+ "kitchen cabinet",
277
+ "clothes",
278
+ "tv",
279
+ "end table",
280
+ "dining table",
281
+ "keyboard",
282
+ "toilet paper",
283
+ "tv stand",
284
+ "whiteboard",
285
+ "trash can",
286
+ "closet",
287
+ "stairs",
288
+ "computer tower",
289
+ "bin",
290
+ "ottoman",
291
+ "washing machine",
292
+ "copier",
293
+ "sofa chair",
294
+ "file cabinet",
295
+ "shower",
296
+ "paper towel dispenser",
297
+ "blinds",
298
+ "suitcase",
299
+ "rail",
300
+ "recycling bin",
301
+ "laundry basket",
302
+ "clothes dryer",
303
+ "toilet paper holder",
304
+ "speaker",
305
+ "bathroom stall",
306
+ "shower wall",
307
+ "cup",
308
+ "storage bin",
309
+ "paper towel roll",
310
+ "bulletin board",
311
+ "kitchen counter",
312
+ "toilet paper dispenser",
313
+ "mini fridge",
314
+ "ball",
315
+ "shower curtain rod",
316
+ "shower door",
317
+ "pillar",
318
+ "ledge",
319
+ "toaster oven",
320
+ "toilet seat cover dispenser",
321
+ "cart",
322
+ "storage container",
323
+ "tissue box",
324
+ "light switch",
325
+ "power outlet",
326
+ "sign",
327
+ "closet door",
328
+ "vacuum cleaner",
329
+ "stuffed animal",
330
+ "headphones",
331
+ "guitar case",
332
+ "hair dryer",
333
+ "water bottle",
334
+ "handicap bar",
335
+ "purse",
336
+ "shower floor",
337
+ "water pitcher",
338
+ "paper bag",
339
+ "projector screen",
340
+ "divider",
341
+ "laundry detergent",
342
+ "bathroom counter",
343
+ "object",
344
+ "bathroom vanity",
345
+ "closet wall",
346
+ "laundry hamper",
347
+ "bathroom stall door",
348
+ "ceiling light",
349
+ "trash bin",
350
+ "stair rail",
351
+ "tube",
352
+ "bathroom cabinet",
353
+ "cd case",
354
+ "closet rod",
355
+ "coffee kettle",
356
+ "structure",
357
+ "shower head",
358
+ "keyboard piano",
359
+ "case of water bottles",
360
+ "coat rack",
361
+ "storage organizer",
362
+ "folded chair",
363
+ "power strip",
364
+ "calendar",
365
+ "poster",
366
+ "potted plant",
367
+ "luggage",
368
+ "mattress",
369
+ ),
370
+ valid_index=(
371
+ (0, 1, 4, 5, 6, 7, 8, 10, 19, 29, 30, 31, 32),
372
+ (0, 1, 2, 3, 4, 5, 6, 7, 8, 11, 13, 14, 15, 16, 17, 18, 19, 20, 21, 23, 25, 26, 33, 34, 35),
373
+ (
374
+ 0,
375
+ 4,
376
+ 36,
377
+ 2,
378
+ 7,
379
+ 1,
380
+ 37,
381
+ 6,
382
+ 8,
383
+ 9,
384
+ 38,
385
+ 39,
386
+ 40,
387
+ 11,
388
+ 19,
389
+ 41,
390
+ 13,
391
+ 42,
392
+ 43,
393
+ 5,
394
+ 25,
395
+ 44,
396
+ 26,
397
+ 45,
398
+ 46,
399
+ 47,
400
+ 3,
401
+ 15,
402
+ 18,
403
+ 48,
404
+ 49,
405
+ 50,
406
+ 51,
407
+ 52,
408
+ 53,
409
+ 54,
410
+ 55,
411
+ 24,
412
+ 56,
413
+ 57,
414
+ 58,
415
+ 59,
416
+ 60,
417
+ 61,
418
+ 62,
419
+ 63,
420
+ 27,
421
+ 22,
422
+ 64,
423
+ 65,
424
+ 66,
425
+ 67,
426
+ 68,
427
+ 69,
428
+ 70,
429
+ 71,
430
+ 72,
431
+ 73,
432
+ 74,
433
+ 75,
434
+ 76,
435
+ 77,
436
+ 78,
437
+ 79,
438
+ 80,
439
+ 81,
440
+ 82,
441
+ 83,
442
+ 84,
443
+ 85,
444
+ 86,
445
+ 87,
446
+ 88,
447
+ 89,
448
+ 90,
449
+ 91,
450
+ 92,
451
+ 93,
452
+ 94,
453
+ 95,
454
+ 96,
455
+ 97,
456
+ 31,
457
+ 98,
458
+ 99,
459
+ 100,
460
+ 101,
461
+ 102,
462
+ 103,
463
+ 104,
464
+ 105,
465
+ 106,
466
+ 107,
467
+ 108,
468
+ 109,
469
+ 110,
470
+ 111,
471
+ 52,
472
+ 112,
473
+ 113,
474
+ 114,
475
+ 115,
476
+ 116,
477
+ 117,
478
+ 118,
479
+ 119,
480
+ 120,
481
+ 121,
482
+ 122,
483
+ 123,
484
+ 124,
485
+ 125,
486
+ 126,
487
+ 127,
488
+ 128,
489
+ 129,
490
+ 130,
491
+ 131,
492
+ 132,
493
+ 133,
494
+ 134,
495
+ 135,
496
+ 136,
497
+ 137,
498
+ 138,
499
+ 139,
500
+ 140,
501
+ 141,
502
+ 142,
503
+ 143,
504
+ 144,
505
+ 145,
506
+ 146,
507
+ 147,
508
+ 148,
509
+ 149,
510
+ 150,
511
+ 151,
512
+ 152,
513
+ 153,
514
+ 154,
515
+ 155,
516
+ 156,
517
+ 157,
518
+ 158,
519
+ 159,
520
+ 160,
521
+ 161,
522
+ 162,
523
+ 163,
524
+ 164,
525
+ 165,
526
+ 166,
527
+ 167,
528
+ 168,
529
+ 169,
530
+ 170,
531
+ 171,
532
+ 172,
533
+ 173,
534
+ 174,
535
+ 175,
536
+ 176,
537
+ 177,
538
+ 178,
539
+ 179,
540
+ 180,
541
+ 181,
542
+ 182,
543
+ 183,
544
+ 184,
545
+ 185,
546
+ 186,
547
+ 187,
548
+ 188,
549
+ 189,
550
+ 190,
551
+ 191,
552
+ 192,
553
+ 193,
554
+ 194,
555
+ 195,
556
+ 196,
557
+ 197,
558
+ 198,
559
+ ),
560
+ (
561
+ 0,
562
+ 4,
563
+ 1,
564
+ 6,
565
+ 7,
566
+ 199,
567
+ 2,
568
+ 200,
569
+ 13,
570
+ 201,
571
+ 3,
572
+ 17,
573
+ 25,
574
+ 11,
575
+ 8,
576
+ 24,
577
+ 9,
578
+ 202,
579
+ 15,
580
+ 36,
581
+ 47,
582
+ 55,
583
+ 40,
584
+ 20,
585
+ 26,
586
+ 203,
587
+ 43,
588
+ 204,
589
+ 205,
590
+ 23,
591
+ 12,
592
+ 16,
593
+ 58,
594
+ 39,
595
+ 48,
596
+ 19,
597
+ 27,
598
+ 206,
599
+ 207,
600
+ 208,
601
+ 52,
602
+ 44,
603
+ 209,
604
+ 68,
605
+ 210,
606
+ 211,
607
+ 77,
608
+ 22,
609
+ 212,
610
+ 213,
611
+ 214,
612
+ 64,
613
+ 71,
614
+ 51,
615
+ 215,
616
+ 53,
617
+ 216,
618
+ 217,
619
+ 75,
620
+ 29,
621
+ 218,
622
+ 18,
623
+ 219,
624
+ 96,
625
+ 220,
626
+ 221,
627
+ 72,
628
+ 67,
629
+ 222,
630
+ 73,
631
+ 94,
632
+ 223,
633
+ 131,
634
+ 224,
635
+ 114,
636
+ 124,
637
+ 86,
638
+ 106,
639
+ 225,
640
+ 226,
641
+ 49,
642
+ 227,
643
+ 92,
644
+ 76,
645
+ 70,
646
+ 61,
647
+ 98,
648
+ 87,
649
+ 74,
650
+ 62,
651
+ 228,
652
+ 123,
653
+ 229,
654
+ 120,
655
+ 230,
656
+ 90,
657
+ 231,
658
+ 31,
659
+ 112,
660
+ 113,
661
+ 232,
662
+ 233,
663
+ 234,
664
+ 63,
665
+ 235,
666
+ 83,
667
+ 101,
668
+ 236,
669
+ 140,
670
+ 89,
671
+ 99,
672
+ 80,
673
+ 116,
674
+ 237,
675
+ 138,
676
+ 142,
677
+ 81,
678
+ 238,
679
+ 41,
680
+ 239,
681
+ 240,
682
+ 121,
683
+ 241,
684
+ 127,
685
+ 242,
686
+ 129,
687
+ 117,
688
+ 115,
689
+ 243,
690
+ 244,
691
+ 245,
692
+ 246,
693
+ 93,
694
+ 247,
695
+ 143,
696
+ 248,
697
+ 249,
698
+ 132,
699
+ 250,
700
+ 251,
701
+ 147,
702
+ 252,
703
+ 84,
704
+ 253,
705
+ 151,
706
+ 254,
707
+ 255,
708
+ 146,
709
+ 118,
710
+ 256,
711
+ 257,
712
+ 110,
713
+ 133,
714
+ 258,
715
+ 85,
716
+ 148,
717
+ 259,
718
+ 260,
719
+ 261,
720
+ 262,
721
+ 145,
722
+ 263,
723
+ 264,
724
+ 111,
725
+ 126,
726
+ 265,
727
+ 137,
728
+ 141,
729
+ 266,
730
+ 267,
731
+ 268,
732
+ 269,
733
+ 270,
734
+ 271,
735
+ 272,
736
+ 273,
737
+ 274,
738
+ 275,
739
+ 276,
740
+ 97,
741
+ 277,
742
+ 278,
743
+ 279,
744
+ 280,
745
+ 281,
746
+ 282,
747
+ 283,
748
+ 284,
749
+ 285,
750
+ 286,
751
+ 287,
752
+ 288,
753
+ 289,
754
+ 139,
755
+ 290,
756
+ 291,
757
+ 292,
758
+ 293,
759
+ 294,
760
+ 295,
761
+ ),
762
+ ),
763
+ backbone_mode=False,
764
+ )
765
+
766
+ # optimizer
767
+ epoch = 800
768
+ eval_epoch = 800
769
+ # epoch = 1600
770
+ # eval_epoch = 1600
771
+ optimizer = dict(type="AdamW", lr=0.005, weight_decay=0.05)
772
+ scheduler = dict(
773
+ type="OneCycleLR",
774
+ max_lr=[0.005, 0.0005],
775
+ pct_start=0.05,
776
+ anneal_strategy="cos",
777
+ div_factor=10.0,
778
+ final_div_factor=1000.0,
779
+ )
780
+ param_dicts = [dict(keyword="block", lr=0.0005)]
781
+
782
+ # datasets
783
+ data = dict(
784
+ num_classes=200,
785
+ ignore_index=-1,
786
+ names=CLASS_LABELS_200,
787
+ train=dict(
788
+ type="ConcatDataset",
789
+ datasets=[
790
+ # Structured3DDataset
791
+ dict(
792
+ type="Structured3DDataset",
793
+ split=["train", "val", "test"],
794
+ data_root="data/structured3d",
795
+ transform=[
796
+ dict(type="CenterShift", apply_z=True),
797
+ dict(type="RandomDropout", dropout_ratio=0.2, dropout_application_ratio=0.2),
798
+ dict(type="RandomRotate", angle=[-1, 1], axis="z", center=[0, 0, 0], p=0.5),
799
+ dict(type="RandomRotate", angle=[-0.015625, 0.015625], axis="x", p=0.5),
800
+ dict(type="RandomRotate", angle=[-0.015625, 0.015625], axis="y", p=0.5),
801
+ dict(type="RandomScale", scale=[0.9, 1.1]),
802
+ dict(type="RandomFlip", p=0.5),
803
+ dict(type="RandomJitter", sigma=0.005, clip=0.02),
804
+ dict(type="ElasticDistortion", distortion_params=[[0.2, 0.4], [0.8, 1.6]]),
805
+ dict(type="ChromaticAutoContrast", p=0.2, blend_factor=None),
806
+ dict(type="ChromaticTranslation", p=0.95, ratio=0.05),
807
+ dict(type="ChromaticJitter", p=0.95, std=0.05),
808
+ dict(type="GridSample", grid_size=0.02, hash_type="fnv", mode="train", return_grid_coord=True),
809
+ dict(type="SphereCrop", sample_rate=0.8, mode="random"),
810
+ dict(type="SphereCrop", point_max=102400, mode="random"),
811
+ dict(type="CenterShift", apply_z=False),
812
+ dict(type="NormalizeColor"),
813
+ dict(type="Add", keys_dict=dict(condition="Structured3D")),
814
+ dict(type="ToTensor"),
815
+ dict(type="Collect", keys=("coord", "grid_coord", "segment", "condition"), feat_keys=("color", "normal")),
816
+ ],
817
+ test_mode=False,
818
+ loop=1,
819
+ ),
820
+ # ScanNet200Dataset
821
+ dict(
822
+ type="ScanNet200Dataset",
823
+ split="train",
824
+ data_root="data/scannet",
825
+ transform=[
826
+ dict(type="CenterShift", apply_z=True),
827
+ dict(type="RandomDropout", dropout_ratio=0.2, dropout_application_ratio=0.2),
828
+ dict(type="RandomRotate", angle=[-1, 1], axis="z", center=[0, 0, 0], p=0.5),
829
+ dict(type="RandomRotate", angle=[-0.015625, 0.015625], axis="x", p=0.5),
830
+ dict(type="RandomRotate", angle=[-0.015625, 0.015625], axis="y", p=0.5),
831
+ dict(type="RandomScale", scale=[0.9, 1.1]),
832
+ dict(type="RandomFlip", p=0.5),
833
+ dict(type="RandomJitter", sigma=0.005, clip=0.02),
834
+ dict(type="ElasticDistortion", distortion_params=[[0.2, 0.4], [0.8, 1.6]]),
835
+ dict(type="ChromaticAutoContrast", p=0.2, blend_factor=None),
836
+ dict(type="ChromaticTranslation", p=0.95, ratio=0.05),
837
+ dict(type="ChromaticJitter", p=0.95, std=0.05),
838
+ dict(type="GridSample", grid_size=0.02, hash_type="fnv", mode="train", return_grid_coord=True),
839
+ dict(type="SphereCrop", point_max=102400, mode="random"),
840
+ dict(type="CenterShift", apply_z=False),
841
+ dict(type="NormalizeColor"),
842
+ dict(type="ShufflePoint"),
843
+ dict(type="Add", keys_dict=dict(condition="ScanNet200")),
844
+ dict(type="ToTensor"),
845
+ dict(type="Collect", keys=("coord", "grid_coord", "segment", "condition"), feat_keys=("color", "normal")),
846
+ ],
847
+ test_mode=False,
848
+ loop=1,
849
+ ),
850
+ # S3DISDataset
851
+ dict(
852
+ type="S3DISDataset",
853
+ split=("Area_1", "Area_2", "Area_3", "Area_4", "Area_6"),
854
+ data_root="data/s3dis",
855
+ transform=[
856
+ dict(type="CenterShift", apply_z=True),
857
+ dict(type="RandomDropout", dropout_ratio=0.2, dropout_application_ratio=0.2),
858
+ dict(type="RandomRotate", angle=[-1, 1], axis="z", center=[0, 0, 0], p=0.5),
859
+ dict(type="RandomRotate", angle=[-0.015625, 0.015625], axis="x", p=0.5),
860
+ dict(type="RandomRotate", angle=[-0.015625, 0.015625], axis="y", p=0.5),
861
+ dict(type="RandomScale", scale=[0.9, 1.1]),
862
+ dict(type="RandomFlip", p=0.5),
863
+ dict(type="RandomJitter", sigma=0.005, clip=0.02),
864
+ dict(type="ChromaticAutoContrast", p=0.2, blend_factor=None),
865
+ dict(type="ChromaticTranslation", p=0.95, ratio=0.05),
866
+ dict(type="ChromaticJitter", p=0.95, std=0.05),
867
+ dict(type="GridSample", grid_size=0.02, hash_type="fnv", mode="train", return_grid_coord=True),
868
+ dict(type="SphereCrop", sample_rate=0.6, mode="random"),
869
+ dict(type="SphereCrop", point_max=204800, mode="random"),
870
+ dict(type="CenterShift", apply_z=False),
871
+ dict(type="NormalizeColor"),
872
+ dict(type="Add", keys_dict=dict(condition="S3DIS")),
873
+ dict(type="ToTensor"),
874
+ dict(type="Collect", keys=("coord", "grid_coord", "segment", "condition"), feat_keys=("color", "normal")),
875
+ ],
876
+ test_mode=False,
877
+ loop=1,
878
+ ),
879
+ # ALC dataset
880
+ dict(
881
+ type="ARKitScenesLabelMakerConsensusDataset",
882
+ split=["train", "val"],
883
+ data_root="data/alc",
884
+ transform=[
885
+ dict(type="CenterShift", apply_z=True),
886
+ dict(type="RandomDropout", dropout_ratio=0.2, dropout_application_ratio=0.2),
887
+ # dict(type="RandomRotateTargetAngle", angle=(1/2, 1, 3/2), center=[0, 0, 0], axis="z", p=0.75),
888
+ dict(type="RandomRotate", angle=[-1, 1], axis="z", center=[0, 0, 0], p=0.5),
889
+ dict(type="RandomRotate", angle=[-1 / 64, 1 / 64], axis="x", p=0.5),
890
+ dict(type="RandomRotate", angle=[-1 / 64, 1 / 64], axis="y", p=0.5),
891
+ dict(type="RandomScale", scale=[0.9, 1.1]),
892
+ # dict(type="RandomShift", shift=[0.2, 0.2, 0.2]),
893
+ dict(type="RandomFlip", p=0.5),
894
+ dict(type="RandomJitter", sigma=0.005, clip=0.02),
895
+ dict(type="ElasticDistortion", distortion_params=[[0.2, 0.4], [0.8, 1.6]]),
896
+ dict(type="ChromaticAutoContrast", p=0.2, blend_factor=None),
897
+ dict(type="ChromaticTranslation", p=0.95, ratio=0.05),
898
+ dict(type="ChromaticJitter", p=0.95, std=0.05),
899
+ # dict(type="HueSaturationTranslation", hue_max=0.2, saturation_max=0.2),
900
+ # dict(type="RandomColorDrop", p=0.2, color_augment=0.0),
901
+ dict(
902
+ type="GridSample",
903
+ grid_size=0.02,
904
+ hash_type="fnv",
905
+ mode="train",
906
+ return_grid_coord=True,
907
+ ),
908
+ dict(type="SphereCrop", point_max=102400, mode="random"),
909
+ dict(type="CenterShift", apply_z=False),
910
+ dict(type="NormalizeColor"),
911
+ # dict(type="ShufflePoint"),
912
+ dict(type="Add", keys_dict=dict(condition="ALC")),
913
+ dict(type="ToTensor"),
914
+ dict(
915
+ type="Collect",
916
+ keys=("coord", "grid_coord", "segment", "condition"),
917
+ feat_keys=("color", "normal"),
918
+ ),
919
+ ],
920
+ test_mode=False,
921
+ ),
922
+ ],
923
+ loop=1,
924
+ ),
925
+ val=dict(
926
+ type="ScanNet200Dataset",
927
+ split="val",
928
+ data_root="data/scannet",
929
+ transform=[
930
+ dict(type="CenterShift", apply_z=True),
931
+ dict(type="GridSample", grid_size=0.02, hash_type="fnv", mode="train", return_grid_coord=True),
932
+ dict(type="CenterShift", apply_z=False),
933
+ dict(type="NormalizeColor"),
934
+ dict(type="ToTensor"),
935
+ dict(type="Add", keys_dict=dict(condition="ScanNet200")),
936
+ dict(type="Collect", keys=("coord", "grid_coord", "segment", "condition"), feat_keys=("color", "normal")),
937
+ ],
938
+ test_mode=False,
939
+ ),
940
+ test=dict(
941
+ type="ScanNet200Dataset",
942
+ split="val",
943
+ data_root="data/scannet",
944
+ transform=[dict(type="CenterShift", apply_z=True), dict(type="NormalizeColor")],
945
+ test_mode=True,
946
+ test_cfg=dict(
947
+ voxelize=dict(type="GridSample", grid_size=0.02, hash_type="fnv", mode="test", keys=("coord", "color", "normal"), return_grid_coord=True),
948
+ crop=None,
949
+ post_transform=[
950
+ dict(type="CenterShift", apply_z=False),
951
+ dict(type="Add", keys_dict=dict(condition="ScanNet200")),
952
+ dict(type="ToTensor"),
953
+ dict(type="Collect", keys=("coord", "grid_coord", "index", "condition"), feat_keys=("color", "normal")),
954
+ ],
955
+ aug_transform=[
956
+ [{"type": "RandomRotateTargetAngle", "angle": [0], "axis": "z", "center": [0, 0, 0], "p": 1}],
957
+ [{"type": "RandomRotateTargetAngle", "angle": [0.5], "axis": "z", "center": [0, 0, 0], "p": 1}],
958
+ [{"type": "RandomRotateTargetAngle", "angle": [1], "axis": "z", "center": [0, 0, 0], "p": 1}],
959
+ [{"type": "RandomRotateTargetAngle", "angle": [1.5], "axis": "z", "center": [0, 0, 0], "p": 1}],
960
+ [{"type": "RandomRotateTargetAngle", "angle": [0], "axis": "z", "center": [0, 0, 0], "p": 1}, {"type": "RandomScale", "scale": [0.95, 0.95]}],
961
+ [{"type": "RandomRotateTargetAngle", "angle": [0.5], "axis": "z", "center": [0, 0, 0], "p": 1}, {"type": "RandomScale", "scale": [0.95, 0.95]}],
962
+ [{"type": "RandomRotateTargetAngle", "angle": [1], "axis": "z", "center": [0, 0, 0], "p": 1}, {"type": "RandomScale", "scale": [0.95, 0.95]}],
963
+ [{"type": "RandomRotateTargetAngle", "angle": [1.5], "axis": "z", "center": [0, 0, 0], "p": 1}, {"type": "RandomScale", "scale": [0.95, 0.95]}],
964
+ [{"type": "RandomRotateTargetAngle", "angle": [0], "axis": "z", "center": [0, 0, 0], "p": 1}, {"type": "RandomScale", "scale": [1.05, 1.05]}],
965
+ [{"type": "RandomRotateTargetAngle", "angle": [0.5], "axis": "z", "center": [0, 0, 0], "p": 1}, {"type": "RandomScale", "scale": [1.05, 1.05]}],
966
+ [{"type": "RandomRotateTargetAngle", "angle": [1], "axis": "z", "center": [0, 0, 0], "p": 1}, {"type": "RandomScale", "scale": [1.05, 1.05]}],
967
+ [{"type": "RandomRotateTargetAngle", "angle": [1.5], "axis": "z", "center": [0, 0, 0], "p": 1}, {"type": "RandomScale", "scale": [1.05, 1.05]}],
968
+ [{"type": "RandomFlip", "p": 1}],
969
+ ],
970
+ ),
971
+ ),
972
+ )
configs/scannetpp/semseg-pt-v3m1-2-ppt-extreme-alc.py ADDED
@@ -0,0 +1,445 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _base_ = [
2
+ "../_base_/default_runtime.py",
3
+ "../_base_/dataset/scannetpp.py",
4
+ ]
5
+
6
+ # misc custom setting
7
+ batch_size = 24 # bs: total bs in all gpus
8
+ num_worker = 48
9
+ mix_prob = 0.8
10
+ empty_cache = False
11
+ enable_amp = True
12
+ find_unused_parameters = True
13
+
14
+ # trainer
15
+ train = dict(
16
+ type="MultiDatasetTrainer",
17
+ )
18
+
19
+ # model settings
20
+ model = dict(
21
+ type="PPT-v1m2",
22
+ backbone=dict(
23
+ type="PT-v3m1",
24
+ in_channels=6,
25
+ order=("z", "z-trans", "hilbert", "hilbert-trans"),
26
+ stride=(2, 2, 2, 2),
27
+ enc_depths=(3, 3, 3, 6, 3),
28
+ enc_channels=(48, 96, 192, 384, 512),
29
+ enc_num_head=(3, 6, 12, 24, 32),
30
+ enc_patch_size=(1024, 1024, 1024, 1024, 1024),
31
+ dec_depths=(3, 3, 3, 3),
32
+ dec_channels=(64, 96, 192, 384),
33
+ dec_num_head=(4, 6, 12, 24),
34
+ dec_patch_size=(1024, 1024, 1024, 1024),
35
+ mlp_ratio=4,
36
+ qkv_bias=True,
37
+ qk_scale=None,
38
+ attn_drop=0.0,
39
+ proj_drop=0.0,
40
+ drop_path=0.3,
41
+ shuffle_orders=True,
42
+ pre_norm=True,
43
+ enable_rpe=False,
44
+ enable_flash=True,
45
+ upcast_attention=False,
46
+ upcast_softmax=False,
47
+ cls_mode=False,
48
+ pdnorm_bn=True,
49
+ pdnorm_ln=True,
50
+ pdnorm_decouple=True,
51
+ pdnorm_adaptive=False,
52
+ pdnorm_affine=True,
53
+ pdnorm_conditions=("ScanNet", "ScanNet++", "S3DIS", "Structured3D", "ALC"),
54
+ ),
55
+ criteria=[
56
+ dict(type="CrossEntropyLoss", loss_weight=1.0, ignore_index=-1),
57
+ dict(type="LovaszLoss", mode="multiclass", loss_weight=1.0, ignore_index=-1),
58
+ ],
59
+ backbone_out_channels=64,
60
+ context_channels=256,
61
+ conditions=("ScanNet", "ScanNet++", "S3DIS", "Structured3D", "ALC"),
62
+ num_classes=(200, 100, 13, 25, 185),
63
+ )
64
+
65
+ # scheduler settings
66
+ epoch = 100
67
+ eval_epoch = 100
68
+ # epoch = 200
69
+ # eval_epoch = 200
70
+ optimizer = dict(type="AdamW", lr=0.005, weight_decay=0.05)
71
+ scheduler = dict(
72
+ type="OneCycleLR",
73
+ max_lr=[0.005, 0.0005],
74
+ pct_start=0.05,
75
+ anneal_strategy="cos",
76
+ div_factor=10.0,
77
+ final_div_factor=1000.0,
78
+ )
79
+ param_dicts = [dict(keyword="block", lr=0.0005)]
80
+
81
+ # dataset settings
82
+ data = dict(
83
+ num_classes=100,
84
+ ignore_index=-1,
85
+ train=dict(
86
+ type="ConcatDataset",
87
+ datasets=[
88
+ # ScanNet
89
+ dict(
90
+ type="ScanNet200Dataset",
91
+ split=["train", "val"],
92
+ data_root="data/scannet",
93
+ transform=[
94
+ dict(type="CenterShift", apply_z=True),
95
+ dict(
96
+ type="RandomDropout",
97
+ dropout_ratio=0.2,
98
+ dropout_application_ratio=0.2,
99
+ ),
100
+ # dict(type="RandomRotateTargetAngle", angle=(1/2, 1, 3/2), center=[0, 0, 0], axis="z", p=0.75),
101
+ dict(
102
+ type="RandomRotate",
103
+ angle=[-1, 1],
104
+ axis="z",
105
+ center=[0, 0, 0],
106
+ p=0.5,
107
+ ),
108
+ dict(type="RandomRotate", angle=[-1 / 64, 1 / 64], axis="x", p=0.5),
109
+ dict(type="RandomRotate", angle=[-1 / 64, 1 / 64], axis="y", p=0.5),
110
+ dict(type="RandomScale", scale=[0.9, 1.1]),
111
+ # dict(type="RandomShift", shift=[0.2, 0.2, 0.2]),
112
+ dict(type="RandomFlip", p=0.5),
113
+ dict(type="RandomJitter", sigma=0.005, clip=0.02),
114
+ dict(
115
+ type="ElasticDistortion",
116
+ distortion_params=[[0.2, 0.4], [0.8, 1.6]],
117
+ ),
118
+ dict(type="ChromaticAutoContrast", p=0.2, blend_factor=None),
119
+ dict(type="ChromaticTranslation", p=0.95, ratio=0.05),
120
+ dict(type="ChromaticJitter", p=0.95, std=0.05),
121
+ # dict(type="HueSaturationTranslation", hue_max=0.2, saturation_max=0.2),
122
+ # dict(type="RandomColorDrop", p=0.2, color_augment=0.0),
123
+ dict(
124
+ type="GridSample",
125
+ grid_size=0.02,
126
+ hash_type="fnv",
127
+ mode="train",
128
+ return_grid_coord=True,
129
+ ),
130
+ dict(type="SphereCrop", point_max=204800, mode="random"),
131
+ dict(type="CenterShift", apply_z=False),
132
+ dict(type="NormalizeColor"),
133
+ dict(type="ShufflePoint"),
134
+ dict(type="Add", keys_dict={"condition": "ScanNet"}),
135
+ dict(type="ToTensor"),
136
+ dict(
137
+ type="Collect",
138
+ keys=("coord", "grid_coord", "segment", "condition"),
139
+ feat_keys=("color", "normal"),
140
+ ),
141
+ ],
142
+ test_mode=False,
143
+ loop=1, # sampling weight
144
+ ),
145
+ # ScanNetPPDataset
146
+ dict(
147
+ type="ScanNetPPDataset",
148
+ # split="train_grid1mm_chunk6x6_stride3x3",
149
+ split=[
150
+ "train_grid1mm_chunk6x6_stride3x3",
151
+ "val_grid1mm_chunk6x6_stride3x3",
152
+ ],
153
+ data_root="data/scannetpp",
154
+ transform=[
155
+ dict(type="CenterShift", apply_z=True),
156
+ dict(
157
+ type="RandomDropout",
158
+ dropout_ratio=0.2,
159
+ dropout_application_ratio=0.2,
160
+ ),
161
+ # dict(type="RandomRotateTargetAngle", angle=(1/2, 1, 3/2), center=[0, 0, 0], axis="z", p=0.75),
162
+ dict(
163
+ type="RandomRotate",
164
+ angle=[-1, 1],
165
+ axis="z",
166
+ center=[0, 0, 0],
167
+ p=0.5,
168
+ ),
169
+ dict(type="RandomRotate", angle=[-1 / 64, 1 / 64], axis="x", p=0.5),
170
+ dict(type="RandomRotate", angle=[-1 / 64, 1 / 64], axis="y", p=0.5),
171
+ dict(type="RandomScale", scale=[0.9, 1.1]),
172
+ # dict(type="RandomShift", shift=[0.2, 0.2, 0.2]),
173
+ dict(type="RandomFlip", p=0.5),
174
+ dict(type="RandomJitter", sigma=0.005, clip=0.02),
175
+ dict(
176
+ type="ElasticDistortion",
177
+ distortion_params=[[0.2, 0.4], [0.8, 1.6]],
178
+ ),
179
+ dict(type="ChromaticAutoContrast", p=0.2, blend_factor=None),
180
+ dict(type="ChromaticTranslation", p=0.95, ratio=0.05),
181
+ dict(type="ChromaticJitter", p=0.95, std=0.05),
182
+ # dict(type="HueSaturationTranslation", hue_max=0.2, saturation_max=0.2),
183
+ # dict(type="RandomColorDrop", p=0.2, color_augment=0.0),
184
+ dict(
185
+ type="GridSample",
186
+ grid_size=0.02,
187
+ hash_type="fnv",
188
+ mode="train",
189
+ return_grid_coord=True,
190
+ ),
191
+ dict(type="SphereCrop", point_max=204800, mode="random"),
192
+ dict(type="CenterShift", apply_z=False),
193
+ dict(type="NormalizeColor"),
194
+ # dict(type="ShufflePoint"),
195
+ dict(type="Add", keys_dict={"condition": "ScanNet++"}),
196
+ dict(type="ToTensor"),
197
+ dict(
198
+ type="Collect",
199
+ keys=("coord", "grid_coord", "segment", "condition"),
200
+ feat_keys=("color", "normal"),
201
+ ),
202
+ ],
203
+ test_mode=False,
204
+ ),
205
+ # ALC dataset
206
+ dict(
207
+ type="ARKitScenesLabelMakerConsensusDataset",
208
+ split=["train", "val"],
209
+ data_root="data/alc",
210
+ transform=[
211
+ dict(type="CenterShift", apply_z=True),
212
+ dict(type="RandomDropout", dropout_ratio=0.2, dropout_application_ratio=0.2),
213
+ # dict(type="RandomRotateTargetAngle", angle=(1/2, 1, 3/2), center=[0, 0, 0], axis="z", p=0.75),
214
+ dict(type="RandomRotate", angle=[-1, 1], axis="z", center=[0, 0, 0], p=0.5),
215
+ dict(type="RandomRotate", angle=[-1 / 64, 1 / 64], axis="x", p=0.5),
216
+ dict(type="RandomRotate", angle=[-1 / 64, 1 / 64], axis="y", p=0.5),
217
+ dict(type="RandomScale", scale=[0.9, 1.1]),
218
+ # dict(type="RandomShift", shift=[0.2, 0.2, 0.2]),
219
+ dict(type="RandomFlip", p=0.5),
220
+ dict(type="RandomJitter", sigma=0.005, clip=0.02),
221
+ dict(type="ElasticDistortion", distortion_params=[[0.2, 0.4], [0.8, 1.6]]),
222
+ dict(type="ChromaticAutoContrast", p=0.2, blend_factor=None),
223
+ dict(type="ChromaticTranslation", p=0.95, ratio=0.05),
224
+ dict(type="ChromaticJitter", p=0.95, std=0.05),
225
+ # dict(type="HueSaturationTranslation", hue_max=0.2, saturation_max=0.2),
226
+ # dict(type="RandomColorDrop", p=0.2, color_augment=0.0),
227
+ dict(
228
+ type="GridSample",
229
+ grid_size=0.02,
230
+ hash_type="fnv",
231
+ mode="train",
232
+ return_grid_coord=True,
233
+ ),
234
+ dict(type="SphereCrop", point_max=102400, mode="random"),
235
+ dict(type="CenterShift", apply_z=False),
236
+ dict(type="NormalizeColor"),
237
+ # dict(type="ShufflePoint"),
238
+ dict(type="Add", keys_dict=dict(condition="ALC")),
239
+ dict(type="ToTensor"),
240
+ dict(
241
+ type="Collect",
242
+ keys=("coord", "grid_coord", "segment", "condition"),
243
+ feat_keys=("color", "normal"),
244
+ ),
245
+ ],
246
+ test_mode=False,
247
+ loop=2,
248
+ ),
249
+ ],
250
+ ),
251
+ val=dict(
252
+ type="ScanNetPPDataset",
253
+ split="val",
254
+ data_root="data/scannetpp",
255
+ transform=[
256
+ dict(type="CenterShift", apply_z=True),
257
+ dict(
258
+ type="GridSample",
259
+ grid_size=0.02,
260
+ hash_type="fnv",
261
+ mode="train",
262
+ return_grid_coord=True,
263
+ ),
264
+ dict(type="CenterShift", apply_z=False),
265
+ dict(type="NormalizeColor"),
266
+ dict(type="ToTensor"),
267
+ dict(type="Add", keys_dict={"condition": "ScanNet++"}),
268
+ dict(
269
+ type="Collect",
270
+ keys=("coord", "grid_coord", "segment", "condition"),
271
+ feat_keys=("color", "normal"),
272
+ ),
273
+ ],
274
+ test_mode=False,
275
+ ),
276
+ test=dict(
277
+ type="ScanNetPPDataset",
278
+ split="test",
279
+ data_root="data/scannetpp",
280
+ transform=[
281
+ dict(type="CenterShift", apply_z=True),
282
+ dict(type="NormalizeColor"),
283
+ dict(type="Copy", keys_dict={"segment": "origin_segment"}),
284
+ dict(
285
+ type="GridSample",
286
+ grid_size=0.01,
287
+ hash_type="fnv",
288
+ mode="train",
289
+ keys=("coord", "color", "normal", "segment"),
290
+ return_inverse=True,
291
+ ),
292
+ ],
293
+ test_mode=True,
294
+ test_cfg=dict(
295
+ voxelize=dict(
296
+ type="GridSample",
297
+ grid_size=0.02,
298
+ hash_type="fnv",
299
+ mode="test",
300
+ keys=("coord", "color", "normal"),
301
+ return_grid_coord=True,
302
+ ),
303
+ crop=None,
304
+ post_transform=[
305
+ dict(type="CenterShift", apply_z=False),
306
+ dict(type="Add", keys_dict={"condition": "ScanNet++"}),
307
+ dict(type="ToTensor"),
308
+ dict(
309
+ type="Collect",
310
+ keys=("coord", "grid_coord", "index", "condition"),
311
+ feat_keys=("color", "normal"),
312
+ ),
313
+ ],
314
+ aug_transform=[
315
+ [
316
+ dict(
317
+ type="RandomRotateTargetAngle",
318
+ angle=[0],
319
+ axis="z",
320
+ center=[0, 0, 0],
321
+ p=1,
322
+ )
323
+ ],
324
+ [
325
+ dict(
326
+ type="RandomRotateTargetAngle",
327
+ angle=[1 / 2],
328
+ axis="z",
329
+ center=[0, 0, 0],
330
+ p=1,
331
+ )
332
+ ],
333
+ [
334
+ dict(
335
+ type="RandomRotateTargetAngle",
336
+ angle=[1],
337
+ axis="z",
338
+ center=[0, 0, 0],
339
+ p=1,
340
+ )
341
+ ],
342
+ [
343
+ dict(
344
+ type="RandomRotateTargetAngle",
345
+ angle=[3 / 2],
346
+ axis="z",
347
+ center=[0, 0, 0],
348
+ p=1,
349
+ )
350
+ ],
351
+ [
352
+ dict(
353
+ type="RandomRotateTargetAngle",
354
+ angle=[0],
355
+ axis="z",
356
+ center=[0, 0, 0],
357
+ p=1,
358
+ ),
359
+ dict(type="RandomScale", scale=[0.95, 0.95]),
360
+ ],
361
+ [
362
+ dict(
363
+ type="RandomRotateTargetAngle",
364
+ angle=[1 / 2],
365
+ axis="z",
366
+ center=[0, 0, 0],
367
+ p=1,
368
+ ),
369
+ dict(type="RandomScale", scale=[0.95, 0.95]),
370
+ ],
371
+ [
372
+ dict(
373
+ type="RandomRotateTargetAngle",
374
+ angle=[1],
375
+ axis="z",
376
+ center=[0, 0, 0],
377
+ p=1,
378
+ ),
379
+ dict(type="RandomScale", scale=[0.95, 0.95]),
380
+ ],
381
+ [
382
+ dict(
383
+ type="RandomRotateTargetAngle",
384
+ angle=[3 / 2],
385
+ axis="z",
386
+ center=[0, 0, 0],
387
+ p=1,
388
+ ),
389
+ dict(type="RandomScale", scale=[0.95, 0.95]),
390
+ ],
391
+ [
392
+ dict(
393
+ type="RandomRotateTargetAngle",
394
+ angle=[0],
395
+ axis="z",
396
+ center=[0, 0, 0],
397
+ p=1,
398
+ ),
399
+ dict(type="RandomScale", scale=[1.05, 1.05]),
400
+ ],
401
+ [
402
+ dict(
403
+ type="RandomRotateTargetAngle",
404
+ angle=[1 / 2],
405
+ axis="z",
406
+ center=[0, 0, 0],
407
+ p=1,
408
+ ),
409
+ dict(type="RandomScale", scale=[1.05, 1.05]),
410
+ ],
411
+ [
412
+ dict(
413
+ type="RandomRotateTargetAngle",
414
+ angle=[1],
415
+ axis="z",
416
+ center=[0, 0, 0],
417
+ p=1,
418
+ ),
419
+ dict(type="RandomScale", scale=[1.05, 1.05]),
420
+ ],
421
+ [
422
+ dict(
423
+ type="RandomRotateTargetAngle",
424
+ angle=[3 / 2],
425
+ axis="z",
426
+ center=[0, 0, 0],
427
+ p=1,
428
+ ),
429
+ dict(type="RandomScale", scale=[1.05, 1.05]),
430
+ ],
431
+ [dict(type="RandomFlip", p=1)],
432
+ ],
433
+ ),
434
+ ),
435
+ )
436
+
437
+ # hook
438
+ hooks = [
439
+ dict(type="CheckpointLoader"),
440
+ dict(type="IterationTimer", warmup_iter=2),
441
+ dict(type="InformationWriter"),
442
+ dict(type="SemSegEvaluator"),
443
+ dict(type="CheckpointSaver", save_freq=None),
444
+ dict(type="PreciseEvaluator", test_last=True),
445
+ ]
pointcept/datasets/alc.py ADDED
@@ -0,0 +1,156 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import glob
2
+ import os
3
+ from collections.abc import Sequence
4
+ from copy import deepcopy
5
+
6
+ import numpy as np
7
+ import torch
8
+ from labelmaker.label_data import get_wordnet
9
+ from torch.utils.data import Dataset
10
+
11
+ from pointcept.utils.cache import shared_dict
12
+ from pointcept.utils.logger import get_root_logger
13
+
14
+ from .builder import DATASETS
15
+ from .preprocessing.alc.preprocess_arkitscenes_labelmaker_consensus import get_wordnet_compact_mapping
16
+ from .preprocessing.scannet.meta_data.scannet200_constants import VALID_CLASS_IDS_20, VALID_CLASS_IDS_200
17
+ from .transform import TRANSFORMS, Compose
18
+
19
+
20
+ @DATASETS.register_module()
21
+ class ARKitScenesLabelMakerConsensusDataset(Dataset):
22
+
23
+ label_key = "semantic_pseudo_gt_wn199"
24
+
25
+ def __init__(
26
+ self,
27
+ split="train",
28
+ data_root="data/alc",
29
+ transform=None,
30
+ ignore_index=-1,
31
+ test_mode=False,
32
+ test_cfg=None,
33
+ cache=False,
34
+ loop=1,
35
+ ):
36
+ super(ARKitScenesLabelMakerConsensusDataset, self).__init__()
37
+ self.get_class_to_id()
38
+
39
+ self.data_root = data_root
40
+ self.split = split
41
+ self.transform = Compose(transform)
42
+ self.cache = cache
43
+ self.loop = loop if not test_mode else 1 # force make loop = 1 while in test mode
44
+ self.test_mode = test_mode
45
+ self.test_cfg = test_cfg if test_mode else None
46
+
47
+ if test_mode:
48
+ self.test_voxelize = TRANSFORMS.build(self.test_cfg.voxelize)
49
+ self.test_crop = TRANSFORMS.build(self.test_cfg.crop) if self.test_cfg.crop else None
50
+ self.post_transform = Compose(self.test_cfg.post_transform)
51
+ self.aug_transform = [Compose(aug) for aug in self.test_cfg.aug_transform]
52
+
53
+ self.data_list = self.get_data_list()
54
+
55
+ self.ignore_index = ignore_index
56
+
57
+ logger = get_root_logger()
58
+ logger.info(
59
+ "Totally {} x {} samples in {} set.".format(
60
+ len(self.data_list),
61
+ self.loop,
62
+ split,
63
+ )
64
+ )
65
+
66
+ def get_class_to_id(self):
67
+ self.class2id = get_wordnet_compact_mapping()[0]
68
+
69
+ def get_data_list(self):
70
+ if isinstance(self.split, str):
71
+ data_list = glob.glob(os.path.join(self.data_root, self.split, "*.pth"))
72
+ elif isinstance(self.split, Sequence):
73
+ data_list = []
74
+ for split in self.split:
75
+ data_list += glob.glob(os.path.join(self.data_root, split, "*.pth"))
76
+ else:
77
+ raise NotImplementedError
78
+ return data_list
79
+
80
+ def get_data(self, idx):
81
+ data_path = self.data_list[idx % len(self.data_list)]
82
+
83
+ if not self.cache:
84
+ data = torch.load(data_path)
85
+ else:
86
+ data_name = data_path.replace(os.path.dirname(self.data_root), "").split(".")[0]
87
+ cache_name = "pointcept" + data_name.replace(os.path.sep, "-")
88
+ data = shared_dict(cache_name)
89
+
90
+ coord = data["coord"]
91
+ color = data["color"]
92
+ normal = data["normal"]
93
+ scene_id = data["scene_id"]
94
+ segment = data[self.label_key].reshape(-1)
95
+ instance = np.ones(coord.shape[0]) * -1
96
+
97
+ data_dict = dict(
98
+ coord=coord,
99
+ normal=normal,
100
+ color=color,
101
+ segment=segment,
102
+ instance=instance,
103
+ scene_id=scene_id,
104
+ )
105
+
106
+ return data_dict
107
+
108
+ def get_data_name(self, idx):
109
+ return os.path.basename(self.data_list[idx % len(self.data_list)]).split(".")[0]
110
+
111
+ def prepare_train_data(self, idx):
112
+ # load data
113
+ data_dict = self.get_data(idx)
114
+ data_dict = self.transform(data_dict)
115
+ return data_dict
116
+
117
+ def prepare_test_data(self, idx):
118
+ # load data
119
+ data_dict = self.get_data(idx)
120
+ segment = data_dict.pop("segment")
121
+ data_dict = self.transform(data_dict)
122
+ data_dict_list = []
123
+ for aug in self.aug_transform:
124
+ data_dict_list.append(aug(deepcopy(data_dict)))
125
+
126
+ input_dict_list = []
127
+ for data in data_dict_list:
128
+ data_part_list = self.test_voxelize(data)
129
+ for data_part in data_part_list:
130
+ if self.test_crop:
131
+ data_part = self.test_crop(data_part)
132
+ else:
133
+ data_part = [data_part]
134
+ input_dict_list += data_part
135
+
136
+ for i in range(len(input_dict_list)):
137
+ input_dict_list[i] = self.post_transform(input_dict_list[i])
138
+ data_dict = dict(fragment_list=input_dict_list, segment=segment, name=self.get_data_name(idx))
139
+ return data_dict
140
+
141
+ def __getitem__(self, idx):
142
+ if self.test_mode:
143
+ return self.prepare_test_data(idx)
144
+ else:
145
+ return self.prepare_train_data(idx)
146
+
147
+ def __len__(self):
148
+ return len(self.data_list) * self.loop
149
+
150
+
151
+ @DATASETS.register_module()
152
+ class ARKitScenesLabelMakerScanNet200Dataset(ARKitScenesLabelMakerConsensusDataset):
153
+ label_key = "semantic_pseudo_gt_scannet200"
154
+
155
+ def get_class_to_id(self):
156
+ self.class2id = np.array(VALID_CLASS_IDS_200)
pointcept/datasets/preprocessing/alc/preprocess_arkitscenes_labelmaker_consensus.py ADDED
@@ -0,0 +1,375 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import warnings
2
+
3
+ import torch
4
+
5
+ warnings.filterwarnings("ignore", category=DeprecationWarning)
6
+
7
+ import argparse
8
+ import glob
9
+ import json
10
+ import multiprocessing as mp
11
+ import os
12
+ from concurrent.futures import ProcessPoolExecutor
13
+ from itertools import repeat
14
+ from pathlib import Path
15
+
16
+ import numpy as np
17
+ import pandas as pd
18
+ import plyfile
19
+ from labelmaker import label_mappings
20
+ from labelmaker.label_data import get_wordnet
21
+ from labelmaker.scannet_200_labels import VALID_CLASS_IDS_200
22
+ from tqdm import tqdm
23
+
24
+ IGNORE_INDEX = -1
25
+
26
+
27
+ def get_wordnet_to_scannet200_mapping():
28
+ table = pd.read_csv(Path(os.path.dirname(os.path.realpath(label_mappings.__file__))) / "mappings" / "label_mapping.csv")
29
+ wordnet = get_wordnet()
30
+ wordnet_keys = [x["name"] for x in wordnet]
31
+ mapping = {}
32
+ for row in table.index:
33
+ if table["wnsynsetkey"][row] not in wordnet_keys:
34
+ continue
35
+ scannet_id = table.loc[row, "id"]
36
+ wordnet199_id = next(x for x in wordnet if x["name"] == table["wnsynsetkey"][row])["id"]
37
+
38
+ if scannet_id in VALID_CLASS_IDS_200:
39
+ mapping.setdefault(wordnet199_id, set()).add(scannet_id)
40
+
41
+ wn199_size = np.array([x["id"] for x in wordnet]).max() + 1
42
+ mapping_array = np.zeros(shape=(wn199_size,), dtype=np.uint16)
43
+ for wordnet199_id in mapping.keys():
44
+ mapping_array[wordnet199_id] = min(mapping[wordnet199_id])
45
+
46
+ return mapping_array
47
+
48
+
49
+ def get_wordnet_compact_mapping():
50
+ wordnet_info = get_wordnet()[1:]
51
+ wordnet_info = sorted(wordnet_info, key=lambda x: x["id"])
52
+
53
+ class2id = np.array([item["id"] for item in wordnet_info])
54
+ id2class = np.array([IGNORE_INDEX] * (class2id.max() + 1))
55
+ for class_, id_ in enumerate(class2id):
56
+ id2class[id_] = class_
57
+
58
+ return class2id, id2class
59
+
60
+
61
+ def get_scannet200_compact_mapping():
62
+ class2id = np.array(VALID_CLASS_IDS_200)
63
+ id2class = np.array([IGNORE_INDEX] * (class2id.max() + 1))
64
+ for class_, id_ in enumerate(VALID_CLASS_IDS_200):
65
+ id2class[id_] = class_
66
+
67
+ return class2id, id2class
68
+
69
+
70
+ def get_wordnet_names():
71
+ wordnet_info = get_wordnet()[1:]
72
+ wordnet_info = sorted(wordnet_info, key=lambda x: x["id"])
73
+
74
+ names = [item["name"].split(".")[0].replace("_", " ") for item in wordnet_info]
75
+
76
+ return names
77
+
78
+
79
+ def read_plypcd(filepath):
80
+ """Read ply file and return it as numpy array. Returns None if emtpy."""
81
+
82
+ with open(filepath, "rb") as f:
83
+ plydata = plyfile.PlyData.read(f)
84
+ if plydata.elements:
85
+ data = plydata.elements[0].data
86
+ coords = np.array([data["x"], data["y"], data["z"]], dtype=np.float32).T
87
+
88
+ colors = None
89
+ if ({"red", "green", "blue"} - set(data.dtype.names)) == set():
90
+ colors = np.array([data["red"], data["green"], data["blue"]], dtype=np.uint8).T
91
+
92
+ normals = None
93
+ if ({"nx", "ny", "nz"} - set(data.dtype.names)) == set():
94
+ normals = np.array([data["nx"], data["ny"], data["nz"]], dtype=np.float32).T
95
+
96
+ return coords, colors, normals
97
+
98
+
99
+ def handle_process(
100
+ scene_dir: str,
101
+ output_path: str,
102
+ label_mapping,
103
+ wn199_id2class,
104
+ scannet200_id2class,
105
+ ):
106
+ scene_dir = Path(scene_dir)
107
+
108
+ print(f"Processing: {scene_dir.name} in {scene_dir.parent.name}")
109
+
110
+ coords, colors, normals = read_plypcd(str(scene_dir / "pcd_downsampled.ply"))
111
+ save_dict = dict(
112
+ coord=coords,
113
+ color=colors,
114
+ scene_id=scene_dir.name,
115
+ normal=normals,
116
+ )
117
+
118
+ label_file = scene_dir / "labels_downsampled.txt"
119
+ wordnet_label = np.loadtxt(str(label_file), dtype=np.uint8).reshape(-1, 1)
120
+ scannet200_label = label_mapping[wordnet_label]
121
+ save_dict["semantic_pseudo_gt_wn199"] = wn199_id2class[wordnet_label]
122
+ save_dict["semantic_pseudo_gt_scannet200"] = scannet200_id2class[scannet200_label]
123
+
124
+ torch.save(save_dict, output_path)
125
+
126
+
127
+ if __name__ == "__main__":
128
+ parser = argparse.ArgumentParser()
129
+ parser.add_argument(
130
+ "--dataset_root",
131
+ required=True,
132
+ help="Path to the ScanNet dataset containing scene folders",
133
+ )
134
+ parser.add_argument(
135
+ "--output_root",
136
+ required=True,
137
+ help="Output path where train/val folders will be located",
138
+ )
139
+ config = parser.parse_args()
140
+
141
+ # Create output directories
142
+ train_output_dir = os.path.join(config.output_root, "train")
143
+ os.makedirs(train_output_dir, exist_ok=True)
144
+ val_output_dir = os.path.join(config.output_root, "val")
145
+ os.makedirs(val_output_dir, exist_ok=True)
146
+
147
+ # Load label map
148
+ wn_scannet200_label_mapping = get_wordnet_to_scannet200_mapping()
149
+ _, wn199_id2class = get_wordnet_compact_mapping()
150
+ _, scannet200_id2class = get_scannet200_compact_mapping()
151
+
152
+ scene_dirs = []
153
+ output_paths = []
154
+
155
+ # Load train/val splits
156
+ train_folder = Path(config.dataset_root) / "Training"
157
+ train_scene_names = os.listdir(str(train_folder))
158
+ for scene in tqdm(train_scene_names):
159
+ file_path = train_folder / scene / "pcd_downsampled.ply"
160
+ if file_path.exists() and os.path.getsize(str(file_path)) <= 50 * 1024 * 1024:
161
+ scene_dirs.append(str(train_folder / scene))
162
+ output_paths.append(str(Path(config.output_root) / "train" / f"{scene}.pth"))
163
+
164
+ val_folder = Path(config.dataset_root) / "Validation"
165
+ val_scene_names = os.listdir(str(val_folder))
166
+ for scene in tqdm(val_scene_names):
167
+ file_path = val_folder / scene / "pcd_downsampled.ply"
168
+ if file_path.exists() and os.path.getsize(str(file_path)) <= 50 * 1024 * 1024:
169
+ scene_dirs.append(str(val_folder / scene))
170
+ output_paths.append(str(Path(config.output_root) / "val" / f"{scene}.pth"))
171
+
172
+ # Preprocess data.
173
+ print("Processing scenes...")
174
+ pool = ProcessPoolExecutor(max_workers=mp.cpu_count())
175
+ print(f"Using {mp.cpu_count()} cores")
176
+ # pool = ProcessPoolExecutor(max_workers=1)
177
+ _ = list(
178
+ pool.map(
179
+ handle_process,
180
+ scene_dirs,
181
+ output_paths,
182
+ repeat(wn_scannet200_label_mapping),
183
+ repeat(wn199_id2class),
184
+ repeat(scannet200_id2class),
185
+ )
186
+ )
187
+
188
+
189
+ WORDNET_NAMES = (
190
+ "wall",
191
+ "chair",
192
+ "book",
193
+ "cabinet",
194
+ "door",
195
+ "floor",
196
+ "ashcan",
197
+ "table",
198
+ "window",
199
+ "bookshelf",
200
+ "display",
201
+ "cushion",
202
+ "box",
203
+ "picture",
204
+ "ceiling",
205
+ "doorframe",
206
+ "desk",
207
+ "swivel chair",
208
+ "towel",
209
+ "sofa",
210
+ "sink",
211
+ "backpack",
212
+ "lamp",
213
+ "chest of drawers",
214
+ "apparel",
215
+ "armchair",
216
+ "bed",
217
+ "curtain",
218
+ "mirror",
219
+ "plant",
220
+ "radiator",
221
+ "toilet tissue",
222
+ "shoe",
223
+ "bag",
224
+ "bottle",
225
+ "countertop",
226
+ "coffee table",
227
+ "toilet",
228
+ "computer keyboard",
229
+ "fridge",
230
+ "stool",
231
+ "computer",
232
+ "mug",
233
+ "telephone",
234
+ "light",
235
+ "jacket",
236
+ "bathtub",
237
+ "shower curtain",
238
+ "microwave",
239
+ "footstool",
240
+ "baggage",
241
+ "laptop",
242
+ "printer",
243
+ "shower stall",
244
+ "soap dispenser",
245
+ "stove",
246
+ "fan",
247
+ "paper",
248
+ "stand",
249
+ "bench",
250
+ "wardrobe",
251
+ "blanket",
252
+ "booth",
253
+ "duplicator",
254
+ "bar",
255
+ "soap dish",
256
+ "switch",
257
+ "coffee maker",
258
+ "decoration",
259
+ "range hood",
260
+ "blackboard",
261
+ "clock",
262
+ "railing",
263
+ "mat",
264
+ "seat",
265
+ "bannister",
266
+ "container",
267
+ "mouse",
268
+ "person",
269
+ "stairway",
270
+ "basket",
271
+ "dumbbell",
272
+ "column",
273
+ "bucket",
274
+ "windowsill",
275
+ "signboard",
276
+ "dishwasher",
277
+ "loudspeaker",
278
+ "washer",
279
+ "paper towel",
280
+ "clothes hamper",
281
+ "piano",
282
+ "sack",
283
+ "handcart",
284
+ "blind",
285
+ "dish rack",
286
+ "mailbox",
287
+ "bag",
288
+ "bicycle",
289
+ "ladder",
290
+ "rack",
291
+ "tray",
292
+ "toaster",
293
+ "paper cutter",
294
+ "plunger",
295
+ "dryer",
296
+ "guitar",
297
+ "fire extinguisher",
298
+ "pitcher",
299
+ "pipe",
300
+ "plate",
301
+ "vacuum",
302
+ "bowl",
303
+ "hat",
304
+ "rod",
305
+ "water cooler",
306
+ "kettle",
307
+ "oven",
308
+ "scale",
309
+ "broom",
310
+ "hand blower",
311
+ "coatrack",
312
+ "teddy",
313
+ "alarm clock",
314
+ "ironing board",
315
+ "fire alarm",
316
+ "machine",
317
+ "music stand",
318
+ "fireplace",
319
+ "furniture",
320
+ "vase",
321
+ "vent",
322
+ "candle",
323
+ "crate",
324
+ "dustpan",
325
+ "earphone",
326
+ "jar",
327
+ "projector",
328
+ "gat",
329
+ "step",
330
+ "step stool",
331
+ "vending machine",
332
+ "coat",
333
+ "coat hanger",
334
+ "drinking fountain",
335
+ "hamper",
336
+ "thermostat",
337
+ "banner",
338
+ "iron",
339
+ "soap",
340
+ "chopping board",
341
+ "kitchen island",
342
+ "shirt",
343
+ "sleeping bag",
344
+ "tire",
345
+ "toothbrush",
346
+ "bathrobe",
347
+ "faucet",
348
+ "slipper",
349
+ "thermos",
350
+ "tripod",
351
+ "dispenser",
352
+ "heater",
353
+ "pool table",
354
+ "remote control",
355
+ "stapler",
356
+ "treadmill",
357
+ "beanbag",
358
+ "dartboard",
359
+ "metronome",
360
+ "rope",
361
+ "sewing machine",
362
+ "shredder",
363
+ "toolbox",
364
+ "water heater",
365
+ "brush",
366
+ "control",
367
+ "dais",
368
+ "dollhouse",
369
+ "envelope",
370
+ "food",
371
+ "frying pan",
372
+ "helmet",
373
+ "tennis racket",
374
+ "umbrella",
375
+ )