groupfree3d_head-L12-O256_4xb8_scannet-seg.py 7.42 KB
Newer Older
hjin2902's avatar
hjin2902 committed
1
_base_ = [
2
3
    '../_base_/datasets/scannet-3d.py', '../_base_/models/groupfree3d.py',
    '../_base_/schedules/schedule-3x.py', '../_base_/default_runtime.py'
hjin2902's avatar
hjin2902 committed
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
]

# model settings
model = dict(
    bbox_head=dict(
        num_classes=18,
        num_decoder_layers=12,
        size_cls_agnostic=False,
        bbox_coder=dict(
            type='GroupFree3DBBoxCoder',
            num_sizes=18,
            num_dir_bins=1,
            with_rot=False,
            size_cls_agnostic=False,
            mean_sizes=[[0.76966727, 0.8116021, 0.92573744],
                        [1.876858, 1.8425595, 1.1931566],
                        [0.61328, 0.6148609, 0.7182701],
                        [1.3955007, 1.5121545, 0.83443564],
                        [0.97949594, 1.0675149, 0.6329687],
                        [0.531663, 0.5955577, 1.7500148],
                        [0.9624706, 0.72462326, 1.1481868],
                        [0.83221924, 1.0490936, 1.6875663],
                        [0.21132214, 0.4206159, 0.5372846],
                        [1.4440073, 1.8970833, 0.26985747],
                        [1.0294262, 1.4040797, 0.87554324],
                        [1.3766412, 0.65521795, 1.6813129],
                        [0.6650819, 0.71111923, 1.298853],
                        [0.41999173, 0.37906948, 1.7513971],
                        [0.59359556, 0.5912492, 0.73919016],
                        [0.50867593, 0.50656086, 0.30136237],
                        [1.1511526, 1.0546296, 0.49706793],
                        [0.47535285, 0.49249494, 0.5802117]]),
        sampling_objectness_loss=dict(
jshilong's avatar
jshilong committed
37
            type='mmdet.FocalLoss',
hjin2902's avatar
hjin2902 committed
38
39
40
41
42
            use_sigmoid=True,
            gamma=2.0,
            alpha=0.25,
            loss_weight=8.0),
        objectness_loss=dict(
jshilong's avatar
jshilong committed
43
            type='mmdet.FocalLoss',
hjin2902's avatar
hjin2902 committed
44
45
46
47
48
            use_sigmoid=True,
            gamma=2.0,
            alpha=0.25,
            loss_weight=1.0),
        center_loss=dict(
jshilong's avatar
jshilong committed
49
50
51
52
            type='mmdet.SmoothL1Loss',
            beta=0.04,
            reduction='sum',
            loss_weight=10.0),
hjin2902's avatar
hjin2902 committed
53
        dir_class_loss=dict(
jshilong's avatar
jshilong committed
54
            type='mmdet.CrossEntropyLoss', reduction='sum', loss_weight=1.0),
hjin2902's avatar
hjin2902 committed
55
        dir_res_loss=dict(
jshilong's avatar
jshilong committed
56
            type='mmdet.SmoothL1Loss', reduction='sum', loss_weight=10.0),
hjin2902's avatar
hjin2902 committed
57
        size_class_loss=dict(
jshilong's avatar
jshilong committed
58
            type='mmdet.CrossEntropyLoss', reduction='sum', loss_weight=1.0),
hjin2902's avatar
hjin2902 committed
59
        size_res_loss=dict(
jshilong's avatar
jshilong committed
60
            type='mmdet.SmoothL1Loss',
hjin2902's avatar
hjin2902 committed
61
62
63
64
            beta=1.0 / 9.0,
            reduction='sum',
            loss_weight=10.0 / 9.0),
        semantic_loss=dict(
jshilong's avatar
jshilong committed
65
            type='mmdet.CrossEntropyLoss', reduction='sum', loss_weight=1.0)),
hjin2902's avatar
hjin2902 committed
66
    test_cfg=dict(
jshilong's avatar
jshilong committed
67
        sample_mode='kps',
hjin2902's avatar
hjin2902 committed
68
69
70
71
72
73
74
75
76
77
78
79
        nms_thr=0.25,
        score_thr=0.0,
        per_class_proposal=True,
        prediction_stages='last_three'))

# dataset settings
dataset_type = 'ScanNetDataset'
data_root = './data/scannet/'
class_names = ('cabinet', 'bed', 'chair', 'sofa', 'table', 'door', 'window',
               'bookshelf', 'picture', 'counter', 'desk', 'curtain',
               'refrigerator', 'showercurtrain', 'toilet', 'sink', 'bathtub',
               'garbagebin')
jshilong's avatar
jshilong committed
80

81
metainfo = dict(classes=class_names)
82
backend_args = None
jshilong's avatar
jshilong committed
83

hjin2902's avatar
hjin2902 committed
84
85
86
87
88
train_pipeline = [
    dict(
        type='LoadPointsFromFile',
        coord_type='DEPTH',
        load_dim=6,
89
90
        use_dim=[0, 1, 2],
        backend_args=backend_args),
hjin2902's avatar
hjin2902 committed
91
92
93
94
95
    dict(
        type='LoadAnnotations3D',
        with_bbox_3d=True,
        with_label_3d=True,
        with_mask_3d=True,
96
97
        with_seg_3d=True,
        backend_args=backend_args),
hjin2902's avatar
hjin2902 committed
98
    dict(type='GlobalAlignment', rotation_axis=2),
99
    dict(type='PointSegClassMapping'),
100
    dict(type='PointSample', num_points=50000),
hjin2902's avatar
hjin2902 committed
101
102
103
104
105
106
107
108
109
110
    dict(
        type='RandomFlip3D',
        sync_2d=False,
        flip_ratio_bev_horizontal=0.5,
        flip_ratio_bev_vertical=0.5),
    dict(
        type='GlobalRotScaleTrans',
        rot_range=[-0.087266, 0.087266],
        scale_ratio_range=[1.0, 1.0]),
    dict(
jshilong's avatar
jshilong committed
111
        type='Pack3DDetInputs',
hjin2902's avatar
hjin2902 committed
112
113
114
115
116
117
118
119
120
121
        keys=[
            'points', 'gt_bboxes_3d', 'gt_labels_3d', 'pts_semantic_mask',
            'pts_instance_mask'
        ])
]
test_pipeline = [
    dict(
        type='LoadPointsFromFile',
        coord_type='DEPTH',
        load_dim=6,
122
123
        use_dim=[0, 1, 2],
        backend_args=backend_args),
hjin2902's avatar
hjin2902 committed
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
    dict(type='GlobalAlignment', rotation_axis=2),
    dict(
        type='MultiScaleFlipAug3D',
        img_scale=(1333, 800),
        pts_scale_ratio=1,
        flip=False,
        transforms=[
            dict(
                type='GlobalRotScaleTrans',
                rot_range=[0, 0],
                scale_ratio_range=[1., 1.],
                translation_std=[0, 0, 0]),
            dict(
                type='RandomFlip3D',
                sync_2d=False,
                flip_ratio_bev_horizontal=0.5,
                flip_ratio_bev_vertical=0.5),
141
            dict(type='PointSample', num_points=50000),
jshilong's avatar
jshilong committed
142
143
        ]),
    dict(type='Pack3DDetInputs', keys=['points'])
hjin2902's avatar
hjin2902 committed
144
145
]

jshilong's avatar
jshilong committed
146
147
148
149
150
train_dataloader = dict(
    batch_size=8,
    num_workers=4,
    sampler=dict(type='DefaultSampler', shuffle=True),
    dataset=dict(
hjin2902's avatar
hjin2902 committed
151
        type='RepeatDataset',
152
        times=5,
hjin2902's avatar
hjin2902 committed
153
154
155
        dataset=dict(
            type=dataset_type,
            data_root=data_root,
jshilong's avatar
jshilong committed
156
            ann_file='scannet_infos_train.pkl',
hjin2902's avatar
hjin2902 committed
157
158
            pipeline=train_pipeline,
            filter_empty_gt=False,
jshilong's avatar
jshilong committed
159
            metainfo=metainfo,
hjin2902's avatar
hjin2902 committed
160
161
            # we use box_type_3d='LiDAR' in kitti and nuscenes dataset
            # and box_type_3d='Depth' in sunrgbd and scannet dataset.
162
163
            box_type_3d='Depth',
            backend_args=backend_args)))
jshilong's avatar
jshilong committed
164
165
166
167
168
val_dataloader = dict(
    batch_size=1,
    num_workers=1,
    sampler=dict(type='DefaultSampler', shuffle=False),
    dataset=dict(
hjin2902's avatar
hjin2902 committed
169
170
        type=dataset_type,
        data_root=data_root,
jshilong's avatar
jshilong committed
171
        ann_file='scannet_infos_val.pkl',
hjin2902's avatar
hjin2902 committed
172
        pipeline=test_pipeline,
jshilong's avatar
jshilong committed
173
        metainfo=metainfo,
hjin2902's avatar
hjin2902 committed
174
        test_mode=True,
175
176
        box_type_3d='Depth',
        backend_args=backend_args))
jshilong's avatar
jshilong committed
177
178
179
180
181
test_dataloader = dict(
    batch_size=1,
    num_workers=1,
    sampler=dict(type='DefaultSampler', shuffle=False),
    dataset=dict(
hjin2902's avatar
hjin2902 committed
182
183
        type=dataset_type,
        data_root=data_root,
jshilong's avatar
jshilong committed
184
        ann_file='scannet_infos_val.pkl',
hjin2902's avatar
hjin2902 committed
185
        pipeline=test_pipeline,
jshilong's avatar
jshilong committed
186
        metainfo=metainfo,
hjin2902's avatar
hjin2902 committed
187
        test_mode=True,
188
189
        box_type_3d='Depth',
        backend_args=backend_args))
jshilong's avatar
jshilong committed
190
191
val_evaluator = dict(type='IndoorMetric')
test_evaluator = val_evaluator
hjin2902's avatar
hjin2902 committed
192
193
194

# optimizer
lr = 0.006
jshilong's avatar
jshilong committed
195
196
197
198
optim_wrapper = dict(
    type='OptimWrapper',
    optimizer=dict(type='AdamW', lr=lr, weight_decay=0.0005),
    clip_grad=dict(max_norm=0.1, norm_type=2),
hjin2902's avatar
hjin2902 committed
199
200
201
202
203
204
205
206
207
208
209
    paramwise_cfg=dict(
        custom_keys={
            'bbox_head.decoder_layers': dict(lr_mult=0.1, decay_mult=1.0),
            'bbox_head.decoder_self_posembeds': dict(
                lr_mult=0.1, decay_mult=1.0),
            'bbox_head.decoder_cross_posembeds': dict(
                lr_mult=0.1, decay_mult=1.0),
            'bbox_head.decoder_query_proj': dict(lr_mult=0.1, decay_mult=1.0),
            'bbox_head.decoder_key_proj': dict(lr_mult=0.1, decay_mult=1.0)
        }))

jshilong's avatar
jshilong committed
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
# learning rate
param_scheduler = [
    dict(
        type='MultiStepLR',
        begin=0,
        end=80,
        by_epoch=True,
        milestones=[56, 68],
        gamma=0.1)
]

# training schedule for 1x
train_cfg = dict(type='EpochBasedTrainLoop', max_epochs=80, val_interval=1)
val_cfg = dict(type='ValLoop')
test_cfg = dict(type='TestLoop')
hjin2902's avatar
hjin2902 committed
225

jshilong's avatar
jshilong committed
226
227
default_hooks = dict(
    checkpoint=dict(type='CheckpointHook', interval=1, max_keep_ckpts=10))