configs/detection3d/bevformer/bevformer_tiny_r50_nuscenes.py

# BEvFormer-tiny consumes at lease 6700M GPU memory
# compared to bevformer_base, bevformer_tiny has
# smaller backbone: R101-DCN -> R50
# smaller BEV: 200*200 -> 50*50
# less encoder layers: 6 -> 3
# smaller input size: 1600*900 -> 800*450
# multi-scale feautres -> single scale features (C5)

_base_ = ['configs/base.py']

# If point cloud range is changed, the models should also change their point
# cloud range accordingly
point_cloud_range = [-51.2, -51.2, -5.0, 51.2, 51.2, 3.0]
voxel_size = [0.2, 0.2, 8]

img_norm_cfg = dict(
    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)

# For nuScenes we usually do 10-class detection
CLASSES = [
    'car', 'truck', 'construction_vehicle', 'bus', 'trailer', 'barrier',
    'motorcycle', 'bicycle', 'pedestrian', 'traffic_cone'
]

input_modality = dict(
    use_lidar=False,
    use_camera=True,
    use_radar=False,
    use_map=False,
    use_external=True)

embed_dim = 256
pos_dim = embed_dim // 2
ffn_dim = embed_dim * 2
num_levels = 1
bev_h = 50
bev_w = 50
queue_length = 3  # each sequence contains `queue_length` frames.

model = dict(
    type='BEVFormer',
    use_grid_mask=True,
    video_test_mode=True,
    extract_feat_serially=True,
    pretrained=dict(img='torchvision://resnet50'),
    img_backbone=dict(
        type='ResNet',
        depth=50,
        num_stages=4,
        # out_indices=(3,),
        out_indices=(4, ),
        frozen_stages=1,
        norm_cfg=dict(type='BN', requires_grad=False),
        norm_eval=True,
        style='pytorch',
        zero_init_residual=True),
    img_neck=dict(
        type='FPN',
        in_channels=[2048],
        out_channels=embed_dim,
        start_level=0,
        add_extra_convs='on_output',
        num_outs=num_levels,
        relu_before_extra_convs=True),
    pts_bbox_head=dict(
        type='BEVFormerHead',
        bev_h=bev_h,
        bev_w=bev_w,
        num_query=900,
        num_classes=10,
        in_channels=embed_dim,
        sync_cls_avg_factor=True,
        with_box_refine=True,
        as_two_stage=False,
        transformer=dict(
            type='PerceptionTransformer',
            rotate_prev_bev=True,
            use_shift=True,
            use_can_bus=True,
            embed_dims=embed_dim,
            encoder=dict(
                type='BEVFormerEncoder',
                num_layers=3,
                pc_range=point_cloud_range,
                num_points_in_pillar=4,
                return_intermediate=False,
                transformerlayers=dict(
                    type='BEVFormerLayer',
                    attn_cfgs=[
                        dict(
                            type='TemporalSelfAttention',
                            embed_dims=embed_dim,
                            num_levels=1),
                        dict(
                            type='SpatialCrossAttention',
                            pc_range=point_cloud_range,
                            deformable_attention=dict(
                                type='MSDeformableAttention3D',
                                embed_dims=embed_dim,
                                num_points=8,
                                num_levels=num_levels),
                            embed_dims=embed_dim,
                        )
                    ],
                    ffn_cfgs=dict(
                        type='FFN',
                        embed_dims=256,
                        feedforward_channels=ffn_dim,
                        num_fcs=2,
                        ffn_drop=0.1,
                        act_cfg=dict(type='ReLU', inplace=True),
                    ),
                    operation_order=('self_attn', 'norm', 'cross_attn', 'norm',
                                     'ffn', 'norm'))),
            decoder=dict(
                type='Detr3DTransformerDecoder',
                num_layers=6,
                return_intermediate=True,
                transformerlayers=dict(
                    type='DetrTransformerDecoderLayer',
                    attn_cfgs=[
                        dict(
                            type='MultiheadAttention',
                            embed_dims=embed_dim,
                            num_heads=8,
                            dropout=0.1),
                        dict(
                            type='CustomMSDeformableAttention',
                            embed_dims=embed_dim,
                            num_levels=1),
                    ],
                    ffn_cfgs=dict(
                        type='FFN',
                        embed_dims=256,
                        feedforward_channels=ffn_dim,
                        num_fcs=2,
                        ffn_drop=0.1,
                        act_cfg=dict(type='ReLU', inplace=True),
                    ),
                    operation_order=('self_attn', 'norm', 'cross_attn', 'norm',
                                     'ffn', 'norm')))),
        bbox_coder=dict(
            type='NMSFreeBBoxCoder',
            post_center_range=[-61.2, -61.2, -10.0, 61.2, 61.2, 10.0],
            pc_range=point_cloud_range,
            max_num=300,
            voxel_size=voxel_size,
            num_classes=10),
        positional_encoding=dict(
            type='LearnedPositionalEncoding',
            num_feats=pos_dim,
            row_num_embed=bev_h,
            col_num_embed=bev_w,
        ),
        loss_cls=dict(
            type='FocalLoss',
            use_sigmoid=True,
            gamma=2.0,
            alpha=0.25,
            loss_weight=2.0),
        loss_bbox=dict(type='L1Loss', loss_weight=0.25),
        loss_iou=dict(type='GIoULoss', loss_weight=0.0)),
    # model training and testing settings
    train_cfg=dict(
        pts=dict(
            grid_size=[512, 512, 1],
            voxel_size=voxel_size,
            point_cloud_range=point_cloud_range,
            out_size_factor=4,
            assigner=dict(
                type='HungarianBBoxAssigner3D',
                cls_cost=dict(type='FocalLossCost', weight=2.0),
                reg_cost=dict(type='BBox3DL1Cost', weight=0.25),
                iou_cost=dict(
                    type='IoUCost', weight=0.0
                ),  # Fake cost. This is just to make it compatible with DETR head.
                pc_range=point_cloud_range))))

dataset_type = 'NuScenesDataset'
data_root = 'data/nuScenes/nuscenes-v1.0/'

train_pipeline = [
    dict(type='RandomScaleImageMultiViewImage', scales=[0.5]),
    dict(type='PhotoMetricDistortionMultiViewImage'),
    dict(type='ObjectRangeFilter', point_cloud_range=point_cloud_range),
    dict(type='ObjectNameFilter', classes=CLASSES),
    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
    # dict(type='RandomScaleImageMultiViewImage', scales=[0.5]),
    dict(type='PadMultiViewImage', size_divisor=32),
    dict(type='DefaultFormatBundle3D', class_names=CLASSES),
    dict(
        type='Collect3D',
        keys=['gt_bboxes_3d', 'gt_labels_3d', 'img'],
        meta_keys=('filename', 'ori_shape', 'img_shape', 'lidar2img',
                   'depth2img', 'cam2img', 'pad_shape', 'scale_factor', 'flip',
                   'pcd_horizontal_flip', 'pcd_vertical_flip', 'box_mode_3d',
                   'box_type_3d', 'img_norm_cfg', 'pcd_trans', 'sample_idx',
                   'prev_idx', 'next_idx', 'pcd_scale_factor', 'pcd_rotation',
                   'pts_filename', 'transformation_3d_flow', 'scene_token',
                   'can_bus'))
]

test_pipeline = [
    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
    dict(
        type='MultiScaleFlipAug3D',
        img_scale=(1600, 900),
        pts_scale_ratio=1,
        flip=False,
        transforms=[
            dict(type='RandomScaleImageMultiViewImage', scales=[0.5]),
            dict(type='PadMultiViewImage', size_divisor=32),
            dict(
                type='DefaultFormatBundle3D',
                class_names=CLASSES,
                with_label=False),
            dict(
                type='Collect3D',
                keys=['img'],
                meta_keys=('filename', 'ori_shape', 'img_shape', 'lidar2img',
                           'depth2img', 'cam2img', 'pad_shape', 'scale_factor',
                           'flip', 'pcd_horizontal_flip', 'pcd_vertical_flip',
                           'box_mode_3d', 'box_type_3d', 'img_norm_cfg',
                           'pcd_trans', 'sample_idx', 'prev_idx', 'next_idx',
                           'pcd_scale_factor', 'pcd_rotation', 'pts_filename',
                           'transformation_3d_flow', 'scene_token', 'can_bus'))
        ])
]

data = dict(
    imgs_per_gpu=1,  # 8gpus, total batch size=8
    workers_per_gpu=4,
    pin_memory=True,
    # shuffler_sampler=dict(type='DistributedGroupSampler'),
    # nonshuffler_sampler=dict(type='DistributedSampler'),
    train=dict(
        type=dataset_type,
        data_source=dict(
            type='Det3dSourceNuScenes',
            data_root=data_root,
            ann_file=data_root + 'nuscenes_infos_temporal_train.pkl',
            pipeline=[
                dict(
                    type='LoadMultiViewImageFromFiles',
                    to_float32=True,
                    backend='turbojpeg'),
                dict(
                    type='LoadAnnotations3D',
                    with_bbox_3d=True,
                    with_label_3d=True,
                    with_attr_label=False)
            ],
            classes=CLASSES,
            modality=input_modality,
            test_mode=False,
            use_valid_flag=True,
            # 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'),
        pipeline=train_pipeline,
        queue_length=queue_length,
    ),
    val=dict(
        imgs_per_gpu=1,
        type=dataset_type,
        data_source=dict(
            type='Det3dSourceNuScenes',
            data_root=data_root,
            ann_file=data_root + 'nuscenes_infos_temporal_val.pkl',
            pipeline=[
                dict(
                    type='LoadMultiViewImageFromFiles',
                    to_float32=True,
                    backend='turbojpeg')
            ],
            classes=CLASSES,
            modality=input_modality,
            test_mode=True),
        pipeline=test_pipeline))

paramwise_cfg = {'img_backbone': dict(lr_mult=0.1)}
optimizer = dict(
    type='AdamW', lr=2e-4, paramwise_options=paramwise_cfg, weight_decay=0.01)

optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
# learning policy
lr_config = dict(
    policy='CosineAnnealing',
    warmup='linear',
    warmup_iters=500,
    warmup_ratio=1.0 / 3,
    min_lr_ratio=1e-3)
total_epochs = 24

eval_config = dict(initial=False, interval=1, gpu_collect=False)
eval_pipelines = [
    dict(
        mode='test',
        data=data['val'],
        dist_eval=True,
        evaluators=[
            dict(
                type='NuScenesEvaluator',
                classes=CLASSES,
                result_names=['pts_bbox'])
        ],
    )
]

log_config = dict(
    interval=50,
    hooks=[dict(type='TextLoggerHook'),
           dict(type='TensorboardLoggerHook')])

checkpoint_config = dict(interval=1)
cudnn_benchmark = True
export = dict(
    export_type='blade',
    blade_config=dict(
        enable_fp16=True,
        fp16_fallback_op_ratio=0.0,
        customize_op_black_list=[
            'aten::select', 'aten::index', 'aten::slice', 'aten::view',
            'aten::upsample', 'aten::clamp'
        ]))