problems in mmdet3d training.

[jin-yiwei]

Thank you for your execllent dataset! I've seen previous issues about low results, and i know you used OpenPCD, I modified the config files of mmdet3d to train pp-lidar according to the previous issues(including point range, voxel sizes and so on). however, I still got very low eval accuracy even ‘0’!!!(i already comented the eval line). I really don't know how to solve this problem, so I can only ask you for help, thank you again!!! (this picture is fisrt epoch)

And this is my config:

_base_ = [
    '../_base_/schedules/cyclic_40e.py',
    '../_base_/default_runtime.py'
]

point_cloud_range = [0, -25.6, -3, 51.2, 25.6, 1]
voxel_size = [0.16, 0.16, 5]
max_points_per_voxel = 10

model = dict(
    type='VoxelNet',
    voxel_layer=dict(
        max_num_points=max_points_per_voxel,
        point_cloud_range=point_cloud_range,
        voxel_size=voxel_size,
        max_voxels=(16000, 40000)
    ),
    voxel_encoder=dict(
        type='PillarFeatureNet',
        in_channels=4,
        feat_channels=[64],
        with_distance=False,
        voxel_size=voxel_size,
        point_cloud_range=point_cloud_range),
    middle_encoder=dict(
        type='PointPillarsScatter', in_channels=64, output_shape=[496, 432]),
    backbone=dict(
        type='SECOND',
        in_channels=64,
        layer_nums=[3, 5, 5],
        layer_strides=[2, 2, 2],
        out_channels=[64, 128, 256]),
    neck=dict(
        type='SECONDFPN',
        in_channels=[64, 128, 256],
        upsample_strides=[1, 2, 4],
        out_channels=[128, 128, 128]),
    bbox_head=dict(
        type='Anchor3DHead',
        num_classes=3,
        in_channels=384,
        feat_channels=384,
        use_direction_classifier=True,
        assign_per_class=True,
        anchor_generator=dict(
            type='AlignedAnchor3DRangeGenerator',
            ranges=[
                [0, -25.6, -0.6, 51.2, 25.6, -0.6],
                [0, -25.6, -0.6, 51.2, 25.6, -0.6],
                [0, -25.6, -1.78, 51.2, 25.6, -1.78],
            ],
            # ranges=[
            #     [0, -39.68, -0.6, 69.12, 39.68, -0.6],
            #     [0, -39.68, -0.6, 69.12, 39.68, -0.6],
            #     [0, -39.68, -1.78, 69.12, 39.68, -1.78],
            #     ],
            sizes=[[0.8, 0.6, 1.73], [1.76, 0.6, 1.73], [3.9, 1.6, 1.56]],
            rotations=[0, 1.57],
            reshape_out=False),
        diff_rad_by_sin=True,
        bbox_coder=dict(type='DeltaXYZWLHRBBoxCoder'),
        loss_cls=dict(
            type='FocalLoss',
            use_sigmoid=True,
            gamma=2.0,
            alpha=0.25,
            loss_weight=1.0),
        loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=2.0),
        loss_dir=dict(
            type='CrossEntropyLoss', use_sigmoid=False, loss_weight=0.2)),
    # model training and testing settings
    train_cfg=dict(
        assigner=[
            dict(  # for Pedestrian
                type='MaxIoUAssigner',
                iou_calculator=dict(type='BboxOverlapsNearest3D'),
                pos_iou_thr=0.5,
                neg_iou_thr=0.35,
                min_pos_iou=0.35,
                ignore_iof_thr=-1),
            dict(  # for Cyclist
                type='MaxIoUAssigner',
                iou_calculator=dict(type='BboxOverlapsNearest3D'),
                pos_iou_thr=0.5,
                neg_iou_thr=0.35,
                min_pos_iou=0.35,
                ignore_iof_thr=-1),
            dict(  # for Car
                type='MaxIoUAssigner',
                iou_calculator=dict(type='BboxOverlapsNearest3D'),
                pos_iou_thr=0.6,
                neg_iou_thr=0.45,
                min_pos_iou=0.45,
                ignore_iof_thr=-1),
        ],
        allowed_border=0,
        pos_weight=-1,
        debug=False),
    test_cfg=dict(
        use_rotate_nms=True,
        nms_across_levels=False,
        nms_thr=0.01,
        score_thr=0.1,
        min_bbox_size=0,
        nms_pre=100,
        max_num=50))

dataset_type = 'KittiDataset'
data_root = "../../vod_lidar/"
class_names = ['Pedestrian', 'Cyclist', 'Car']
input_modality = dict(use_lidar=True, use_camera=True)
file_client_args = dict(backend='disk')

db_sampler = dict(
    data_root=data_root,
    info_path=data_root + 'vod_dbinfos_train.pkl',
    rate=1.0,
    prepare=dict(
        filter_by_difficulty=[-1],
        filter_by_min_points=dict(Car=5, Pedestrian=5, Cyclist=5)),
    classes=class_names,
    sample_groups=dict(Car=15, Pedestrian=15, Cyclist=15),
    points_loader=dict(
        type='LoadPointsFromFile',
        coord_type='LIDAR',
        load_dim=4,
        use_dim=4,
        file_client_args=file_client_args),
    file_client_args=file_client_args)

train_pipeline = [
    dict(
        type='LoadPointsFromFile',
        coord_type='LIDAR',
        load_dim=4,
        use_dim=4,
        file_client_args=file_client_args),
    dict(
        type='LoadAnnotations3D',
        with_bbox_3d=True,
        with_label_3d=True,
        file_client_args=file_client_args),
    dict(type='ObjectSample', db_sampler=db_sampler, use_ground_plane=False),
    dict(type='RandomFlip3D', flip_ratio_bev_horizontal=0.5),
    dict(
        type='GlobalRotScaleTrans',
        rot_range=[-0.78539816, 0.78539816],
        scale_ratio_range=[0.95, 1.05]),
    dict(type='PointsRangeFilter', point_cloud_range=point_cloud_range),
    dict(type='ObjectRangeFilter', point_cloud_range=point_cloud_range),
    dict(type='PointShuffle'),
    dict(type='DefaultFormatBundle3D', class_names=class_names),
    dict(type='Collect3D', keys=['points', 'gt_bboxes_3d', 'gt_labels_3d'])
]
test_pipeline = [
    dict(
        type='LoadPointsFromFile',
        coord_type='LIDAR',
        load_dim=4,
        use_dim=4,
        file_client_args=file_client_args),
    dict(
        type='MultiScaleFlipAug3D',
        img_scale=(1333, 800),
        pts_scale_ratio=1,
        flip=False,
        transforms=[
            dict(
                type='GlobalRotScaleTrans',
                rot_range=[-0.78539816, 0.78539816],
                scale_ratio_range=[0.95, 1.05]),
            dict(type='RandomFlip3D'),
            dict(type='PointsRangeFilter', point_cloud_range=point_cloud_range),
            dict(
                type='DefaultFormatBundle3D',
                class_names=class_names,
                with_label=False),
            dict(type='Collect3D', keys=['points'])
        ])
]

eval_pipeline = [
    dict(
        type='LoadPointsFromFile',
        coord_type='LIDAR',
        load_dim=4,
        use_dim=4,
        file_client_args=file_client_args),
    dict(
        type='DefaultFormatBundle3D',
        class_names=class_names,
        with_label=False),
    dict(type='Collect3D', keys=['points'])
]

data = dict(
    samples_per_gpu=12,
    workers_per_gpu=4,
    train=dict(
        type='RepeatDataset',
        times=2,
        dataset=dict(
            type=dataset_type,
            data_root=data_root,
            ann_file=data_root + 'vod_infos_train.pkl',
            split='training',
            pts_prefix='velodyne_reduced',
            pipeline=train_pipeline,
            modality=input_modality,
            classes=class_names,
            test_mode=False,
            # we use box_type_3d='LiDAR' in kitti and nuscenes dataset
            # and box_type_3d='Depth' in sunrgbd and scannet dataset.
            box_type_3d='LiDAR',
            file_client_args=file_client_args)),
    val=dict(
        type=dataset_type,
        data_root=data_root,
        ann_file=data_root + 'vod_infos_val.pkl',
        split='training',
        pts_prefix='velodyne_reduced',
        pipeline=test_pipeline,
        modality=input_modality,
        classes=class_names,
        test_mode=True,
        box_type_3d='LiDAR',
        file_client_args=file_client_args),
    test=dict(
        type=dataset_type,
        data_root=data_root,
        ann_file=data_root + 'vod_infos_val.pkl',
        split='training',
        pts_prefix='velodyne_reduced',
        pipeline=test_pipeline,
        modality=input_modality,
        classes=class_names,
        test_mode=True,
        box_type_3d='LiDAR',
        file_client_args=file_client_args)
)

lr = 0.001
optimizer = dict(lr=lr)
optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
runner = dict(max_epochs=50)
evaluation = dict(interval=1)

[andraspalffy]

As discussed before and in with emails with you, we do not provide support for your implementations for object detection.

Some clear issues that I can see here though, and can help others:

• voxel is taller than the point cloud range height you use
• max number of points per voxel is set to 10, too low for LiDAR
• you use the built in eval tool, not ours.

Hope these helps, also have a look please at previous issues.

Good luck!

[ME-SJTU]

@vehicle-jyw hello, I also applied this dataset to mmdetection3d. Have you solved this issue?

[LiuLiu-55]

@vehicle-jyw hello, I also meet this problem. Have you solved this issue?

[Lime2333]

@LiuLiu-55 @JZhong-Li hello,when I use the dataset to mmdet3d, I encountered a few problems. Have you solved this issue?

[LiuLiu-55]

@LiuLiu-55 @JZhong-Li hello,when I use the dataset to mmdet3d, I encountered a few problems. Have you solved this issue? I use the eval tool they provide not the OpenPCDet, and the performance gets better.

[Lime2333]

@LiuLiu-55 @JZhong-Li hello,when I use the dataset to mmdet3d, I encountered a few problems. Have you solved this issue? I use the eval tool they provide not the OpenPCDet, and the performance gets better.

@LiuLiu-55 Then do you finally reappear at the results of the Pointpillars reappeared at MMDET3D.