td3d_is_s3dis-3d-13class_pretrain.py

voxel_size = .02
padding = .08
n_points = 100000

class_names = ('ceiling', 'floor', 'wall', 'beam', 'column', 'window', 'door',
               'table', 'chair', 'sofa', 'bookcase', 'board', 'clutter')

model = dict(
    type='TD3DInstanceSegmentor',
    voxel_size=voxel_size,
    backbone=dict(type='MinkResNet', in_channels=3, depth=34, norm='batch', return_stem=True, stride=1),
    neck=dict(
        type='NgfcTinySegmentationNeck',
        in_channels=(64, 128, 256, 512),
        out_channels=128),
    head=dict(
        type='TD3DInstanceHead',
        in_channels=128,
        n_reg_outs=6,
        n_classes=len(class_names),
        n_levels=4,
        padding=padding,
        voxel_size=voxel_size,
        unet=dict(
            type='MinkUNet14B', 
            in_channels=32, 
            out_channels=len(class_names) + 1,
            D=3),
        first_assigner=dict(
            type='S3DISAssigner',
            top_pts_threshold=6,
            label2level=[3, 3, 3, 3, 2, 2, 2, 2, 1, 2, 2, 1, 1]),
        second_assigner=dict(
            type='MaxIoU3DAssigner',
            threshold=.25),
        roi_extractor=dict(
            type='Mink3DRoIExtractor',
            voxel_size=voxel_size,
            padding=padding,
            min_pts_threshold=10)),
    train_cfg=dict(num_rois=1),
    test_cfg=dict(
        nms_pre=700,
        iou_thr=.2,
        score_thr=.05,
        binary_score_thr=0.2))

optimizer = dict(type='AdamW', lr=0.001, weight_decay=0.0001)
optimizer_config = dict(grad_clip=dict(max_norm=10, norm_type=2))
lr_config = dict(policy='step', warmup=None, step=[28, 32])
runner = dict(type='EpochBasedRunner', max_epochs=33)
custom_hooks = [dict(type='EmptyCacheHook', after_iter=True)]

checkpoint_config = dict(interval=1, max_keep_ckpts=50)
log_config = dict(
    interval=50,
    hooks=[
        dict(type='TextLoggerHook'),
        # dict(type='TensorboardLoggerHook')
])
dist_params = dict(backend='nccl')
log_level = 'INFO'
work_dir = None
load_from = "td3d_scannet.pth"
resume_from = None
workflow = [('train', 1)]

dataset_type = 'S3DISInstanceSegDataset'
data_root = './data/s3dis/'


train_area = [1, 2, 3, 4, 6]
test_area = 5

train_pipeline = [
    dict(
        type='LoadPointsFromFile',
        coord_type='DEPTH',
        shift_height=False,
        use_color=True,
        load_dim=6,
        use_dim=[0, 1, 2, 3, 4, 5]),
    dict(
        type='LoadAnnotations3D',
        with_mask_3d=True,
        with_seg_3d=True),
    dict(type='PointSample', num_points=n_points),
    dict(type='PointSegClassMappingV2',
        valid_cat_ids=tuple(range(len(class_names))),
        max_cat_id=13),
    dict(
        type='RandomFlip3D',
        sync_2d=False,
        flip_ratio_bev_horizontal=0.5,
        flip_ratio_bev_vertical=0.5),
    dict(
        type='GlobalRotScaleTrans',
        rot_range=[0, 0],
        scale_ratio_range=[0.95, 1.05],
        translation_std=[.1, .1, .1],
        shift_height=False),
    dict(
        type='BboxRecalculation'),
    dict(type='NormalizePointsColor', color_mean=None),
    dict(type='DefaultFormatBundle3D', class_names=class_names),
    dict(type='Collect3D', keys=['points', 'gt_bboxes_3d', 'gt_labels_3d',
                                 'pts_semantic_mask', 'pts_instance_mask'])
]
test_pipeline = [
    dict(
        type='LoadPointsFromFile',
        coord_type='DEPTH',
        shift_height=False,
        use_color=True,
        load_dim=6,
        use_dim=[0, 1, 2, 3, 4, 5]),
    dict(
        type='MultiScaleFlipAug3D',
        img_scale=(1333, 800),
        pts_scale_ratio=1,
        flip=False,
        transforms=[
            dict(type='NormalizePointsColor', color_mean=None),
            dict(
                type='DefaultFormatBundle3D',
                class_names=class_names,
                with_label=False),
            dict(type='Collect3D', keys=['points'])
        ])
]
data = dict(
    samples_per_gpu=4,
    workers_per_gpu=6,
    train=dict(
        type='RepeatDataset',
        times=13,
        dataset=dict(
            type='ConcatDataset',
            datasets=[
                dict(
                    type=dataset_type,
                    data_root=data_root,
                    ann_file=data_root + f's3dis_infos_Area_{i}.pkl',
                    pipeline=train_pipeline,
                    filter_empty_gt=True,
                    classes=class_names,
                    box_type_3d='Depth') for i in train_area
            ],
            separate_eval=False)),
    val=dict(
        type=dataset_type,
        data_root=data_root,
        ann_file=data_root + f's3dis_infos_Area_{test_area}.pkl',
        pipeline=test_pipeline,
        filter_empty_gt=False,
        classes=class_names,
        test_mode=True,
        box_type_3d='Depth'),
    test=dict(
        type=dataset_type,
        data_root=data_root,
        ann_file=data_root + f's3dis_infos_Area_{test_area}.pkl',
        pipeline=test_pipeline,
        filter_empty_gt=False,
        classes=class_names,
        test_mode=True,
        box_type_3d='Depth'))