Problems when replicating the result using nuscenes-mini

[JohnFuguiWang]

Problem description

I tried to train the model using nuscenes-mini dataset, and found the result statistics were quite low (around 0.45). Another weird thing was that the loss declined gradually, but the object map stayed at the same level of 0.45 from the very beginning to the end, and it even went downward with training going on. By the way, I've checked issue #463 and modified the sweeps_num as 9, and the result above was already the result after this modification. So did anyone also use nusenes-mini and how was your result? And why did the map figure stay at the same level during training? Many thanks for your help!

To replicate the problem

I use nuscenes-mini dataset, set the max_epochs as 15, both samples_per_gpu and workers_per_gpu as 2, and use the official order of detect: "torchpack dist-run -np 2 python tools/train.py configs/nuscenes/det/transfusion/secfpn/camera+lidar/swint_v0p075/convfuser.yaml --model.encoders.camera.backbone.init_cfg.checkpoint pretrained/swint-nuimages-pretrained.pth --load_from pretrained/lidar-only-det.pth"

More details

I can also attach some visualization results and the complete config.yaml here if you want to see more details. The content of config.yaml can be found in the next comment of this issue.

[JohnFuguiWang]

Here is the content of my config.yaml if you wanna have a look at: augment2d: gridmask: fixed_prob: true prob: 0.0 resize:

• • 0.38
  • 0.55
• • 0.48
  • 0.48 rotate:
• -5.4
• 5.4 augment3d: rotate:
• -0.78539816
• 0.78539816 scale:
• 0.9
• 1.1 translate: 0.5 checkpoint_config: interval: 1 max_keep_ckpts: 1 cudnn_benchmark: false data: samples_per_gpu: 2 test: ann_file: data/nuscenes/nuscenes_infos_val.pkl box_type_3d: LiDAR dataset_root: data/nuscenes/ map_classes:
  • drivable_area
  • ped_crossing
  • walkway
  • stop_line
  • carpark_area
  • divider modality: use_camera: true use_external: false use_lidar: true use_map: false use_radar: false object_classes:
  • car
  • truck
  • construction_vehicle
  • bus
  • trailer
  • barrier
  • motorcycle
  • bicycle
  • pedestrian
  • traffic_cone pipeline:
  • to_float32: true type: LoadMultiViewImageFromFiles
  • coord_type: LIDAR load_augmented: null load_dim: 5 reduce_beams: 32 type: LoadPointsFromFile use_dim: 5
  • load_augmented: null load_dim: 5 pad_empty_sweeps: true reduce_beams: 32 remove_close: true sweeps_num: 9 type: LoadPointsFromMultiSweeps use_dim: 5
  • compensate_velocity: true filtering: none load_dim: 18 max_num: 2500 normalize: false sweeps_num: 6 type: LoadRadarPointsMultiSweeps use_dim:
    • 0
    • 1
    • 2
    • 5
    • 8
    • 9
    • 18
    • 19
    • 20
    • 21
    • 22
    • 23
    • 24
    • 25
    • 26
    • 27
    • 28
    • 29
    • 30
    • 31
    • 32
    • 33
    • 34
    • 35
    • 36
    • 37
    • 38
    • 39
    • 40
    • 41
    • 42
    • 43
    • 44
    • 45
    • 46
    • 47
    • 48
    • 49
    • 50
    • 51
    • 52
    • 53
    • 54
    • 55
    • 56
  • type: LoadAnnotations3D with_attr_label: false with_bbox_3d: true with_label_3d: true
  • bot_pct_lim:
    • 0.0
    • 0.0 final_dim:
    • 256
    • 704 is_train: false rand_flip: false resize_lim:
    • 0.48
    • 0.48 rot_lim:
    • 0.0
    • 0.0 type: ImageAug3D
  • is_train: false resize_lim:
    • 1.0
    • 1.0 rot_lim:
    • 0.0
    • 0.0 trans_lim: 0.0 type: GlobalRotScaleTrans
  • point_cloud_range:
    • -54.0
    • -54.0
    • -5.0
    • 54.0
    • 54.0
    • 3.0 type: PointsRangeFilter
  • mean:
    • 0.485
    • 0.456
    • 0.406 std:
    • 0.229
    • 0.224
    • 0.225 type: ImageNormalize
  • classes:
    • car
    • truck
    • construction_vehicle
    • bus
    • trailer
    • barrier
    • motorcycle
    • bicycle
    • pedestrian
    • traffic_cone type: DefaultFormatBundle3D
  • keys:
    • img
    • points
    • radar
    • gt_bboxes_3d
    • gt_labels_3d meta_keys:
    • camera_intrinsics
    • camera2ego
    • lidar2ego
    • lidar2camera
    • lidar2image
    • camera2lidar
    • img_aug_matrix
    • lidar_aug_matrix type: Collect3D
  • keyframe_only: true type: GTDepth test_mode: true type: NuScenesDataset train: dataset: ann_file: data/nuscenes/nuscenes_infos_train.pkl box_type_3d: LiDAR dataset_root: data/nuscenes/ map_classes:
    • drivable_area
    • ped_crossing
    • walkway
    • stop_line
    • carpark_area
    • divider modality: use_camera: true use_external: false use_lidar: true use_map: false use_radar: false object_classes:
    • car
    • truck
    • construction_vehicle
    • bus
    • trailer
    • barrier
    • motorcycle
    • bicycle
    • pedestrian
    • traffic_cone pipeline:
    • to_float32: true type: LoadMultiViewImageFromFiles
    • coord_type: LIDAR load_augmented: null load_dim: 5 reduce_beams: 32 type: LoadPointsFromFile use_dim: 5
    • load_augmented: null load_dim: 5 pad_empty_sweeps: true reduce_beams: 32 remove_close: true sweeps_num: 9 type: LoadPointsFromMultiSweeps use_dim: 5
    • compensate_velocity: true filtering: none load_dim: 18 max_num: 2500 normalize: false sweeps_num: 6 type: LoadRadarPointsMultiSweeps use_dim:
    • 0
    • 1
    • 2
    • 5
    • 8
    • 9
    • 18
    • 19
    • 20
    • 21
    • 22
    • 23
    • 24
    • 25
    • 26
    • 27
    • 28
    • 29
    • 30
    • 31
    • 32
    • 33
    • 34
    • 35
    • 36
    • 37
    • 38
    • 39
    • 40
    • 41
    • 42
    • 43
    • 44
    • 45
    • 46
    • 47
    • 48
    • 49
    • 50
    • 51
    • 52
    • 53
    • 54
    • 55
    • 56
    • type: LoadAnnotations3D with_attr_label: false with_bbox_3d: true with_label_3d: true
    • db_sampler: classes:
      • car
      • truck
      • construction_vehicle
      • bus
      • trailer
      • barrier
      • motorcycle
      • bicycle
      • pedestrian
      • traffic_cone dataset_root: data/nuscenes/ info_path: data/nuscenes/nuscenes_dbinfos_train.pkl points_loader: coord_type: LIDAR load_dim: 5 reduce_beams: 32 type: LoadPointsFromFile use_dim: 5 prepare: filter_by_difficulty:
      • -1 filter_by_min_points: barrier: 5 bicycle: 5 bus: 5 car: 5 construction_vehicle: 5 motorcycle: 5 pedestrian: 5 traffic_cone: 5 trailer: 5 truck: 5 rate: 1.0 sample_groups: barrier: 2 bicycle: 6 bus: 4 car: 2 construction_vehicle: 7 motorcycle: 6 pedestrian: 2 traffic_cone: 2 trailer: 6 truck: 3 stop_epoch: -1 type: ObjectPaste
    • bot_pct_lim:
    • 0.0
    • 0.0 final_dim:
    • 256
    • 704 is_train: true rand_flip: true resize_lim:
    • 0.38
    • 0.55 rot_lim:
    • -5.4
    • 5.4 type: ImageAug3D
    • is_train: true resize_lim:
    • 0.9
    • 1.1 rot_lim:
    • -0.78539816
    • 0.78539816 trans_lim: 0.5 type: GlobalRotScaleTrans
    • type: RandomFlip3D
    • point_cloud_range:
    • -54.0
    • -54.0
    • -5.0
    • 54.0
    • 54.0
    • 3.0 type: PointsRangeFilter
    • point_cloud_range:
    • -54.0
    • -54.0
    • -5.0
    • 54.0
    • 54.0
    • 3.0 type: ObjectRangeFilter
    • classes:
    • car
    • truck
    • construction_vehicle
    • bus
    • trailer
    • barrier
    • motorcycle
    • bicycle
    • pedestrian
    • traffic_cone type: ObjectNameFilter
    • mean:
    • 0.485
    • 0.456
    • 0.406 std:
    • 0.229
    • 0.224
    • 0.225 type: ImageNormalize
    • fixed_prob: true max_epoch: 15 mode: 1 offset: false prob: 0.0 ratio: 0.5 rotate: 1 type: GridMask use_h: true use_w: true
    • type: PointShuffle
    • classes:
    • car
    • truck
    • construction_vehicle
    • bus
    • trailer
    • barrier
    • motorcycle
    • bicycle
    • pedestrian
    • traffic_cone type: DefaultFormatBundle3D
    • keys:
    • img
    • points
    • radar
    • gt_bboxes_3d
    • gt_labels_3d meta_keys:
    • camera_intrinsics
    • camera2ego
    • lidar2ego
    • lidar2camera
    • lidar2image
    • camera2lidar
    • img_aug_matrix
    • lidar_aug_matrix type: Collect3D
    • keyframe_only: true type: GTDepth test_mode: false type: NuScenesDataset use_valid_flag: true type: CBGSDataset val: ann_file: data/nuscenes/nuscenes_infos_val.pkl box_type_3d: LiDAR dataset_root: data/nuscenes/ map_classes:
  • drivable_area
  • ped_crossing
  • walkway
  • stop_line
  • carpark_area
  • divider modality: use_camera: true use_external: false use_lidar: true use_map: false use_radar: false object_classes:
  • car
  • truck
  • construction_vehicle
  • bus
  • trailer
  • barrier
  • motorcycle
  • bicycle
  • pedestrian
  • traffic_cone pipeline:
  • to_float32: true type: LoadMultiViewImageFromFiles
  • coord_type: LIDAR load_augmented: null load_dim: 5 reduce_beams: 32 type: LoadPointsFromFile use_dim: 5
  • load_augmented: null load_dim: 5 pad_empty_sweeps: true reduce_beams: 32 remove_close: true sweeps_num: 9 type: LoadPointsFromMultiSweeps use_dim: 5
  • compensate_velocity: true filtering: none load_dim: 18 max_num: 2500 normalize: false sweeps_num: 6 type: LoadRadarPointsMultiSweeps use_dim:
    • 0
    • 1
    • 2
    • 5
    • 8
    • 9
    • 18
    • 19
    • 20
    • 21
    • 22
    • 23
    • 24
    • 25
    • 26
    • 27
    • 28
    • 29
    • 30
    • 31
    • 32
    • 33
    • 34
    • 35
    • 36
    • 37
    • 38
    • 39
    • 40
    • 41
    • 42
    • 43
    • 44
    • 45
    • 46
    • 47
    • 48
    • 49
    • 50
    • 51
    • 52
    • 53
    • 54
    • 55
    • 56
  • type: LoadAnnotations3D with_attr_label: false with_bbox_3d: true with_label_3d: true
  • bot_pct_lim:
    • 0.0
    • 0.0 final_dim:
    • 256
    • 704 is_train: false rand_flip: false resize_lim:
    • 0.48
    • 0.48 rot_lim:
    • 0.0
    • 0.0 type: ImageAug3D
  • is_train: false resize_lim:
    • 1.0
    • 1.0 rot_lim:
    • 0.0
    • 0.0 trans_lim: 0.0 type: GlobalRotScaleTrans
  • point_cloud_range:
    • -54.0
    • -54.0
    • -5.0
    • 54.0
    • 54.0
    • 3.0 type: PointsRangeFilter
  • mean:
    • 0.485
    • 0.456
    • 0.406 std:
    • 0.229
    • 0.224
    • 0.225 type: ImageNormalize
  • classes:
    • car
    • truck
    • construction_vehicle
    • bus
    • trailer
    • barrier
    • motorcycle
    • bicycle
    • pedestrian
    • traffic_cone type: DefaultFormatBundle3D
  • keys:
    • img
    • points
    • radar
    • gt_bboxes_3d
    • gt_labels_3d meta_keys:
    • camera_intrinsics
    • camera2ego
    • lidar2ego
    • lidar2camera
    • lidar2image
    • camera2lidar
    • img_aug_matrix
    • lidar_aug_matrix type: Collect3D
  • keyframe_only: true type: GTDepth test_mode: false type: NuScenesDataset workers_per_gpu: 2 dataset_root: data/nuscenes/ dataset_type: NuScenesDataset deterministic: false evaluation: interval: 1 pipeline:
• to_float32: true type: LoadMultiViewImageFromFiles
• coord_type: LIDAR load_augmented: null load_dim: 5 reduce_beams: 32 type: LoadPointsFromFile use_dim: 5
• load_augmented: null load_dim: 5 pad_empty_sweeps: true reduce_beams: 32 remove_close: true sweeps_num: 9 type: LoadPointsFromMultiSweeps use_dim: 5
• compensate_velocity: true filtering: none load_dim: 18 max_num: 2500 normalize: false sweeps_num: 6 type: LoadRadarPointsMultiSweeps use_dim:
  • 0
  • 1
  • 2
  • 5
  • 8
  • 9
  • 18
  • 19
  • 20
  • 21
  • 22
  • 23
  • 24
  • 25
  • 26
  • 27
  • 28
  • 29
  • 30
  • 31
  • 32
  • 33
  • 34
  • 35
  • 36
  • 37
  • 38
  • 39
  • 40
  • 41
  • 42
  • 43
  • 44
  • 45
  • 46
  • 47
  • 48
  • 49
  • 50
  • 51
  • 52
  • 53
  • 54
  • 55
  • 56
• type: LoadAnnotations3D with_attr_label: false with_bbox_3d: true with_label_3d: true
• bot_pct_lim:
  • 0.0
  • 0.0 final_dim:
  • 256
  • 704 is_train: false rand_flip: false resize_lim:
  • 0.48
  • 0.48 rot_lim:
  • 0.0
  • 0.0 type: ImageAug3D
• is_train: false resize_lim:
  • 1.0
  • 1.0 rot_lim:
  • 0.0
  • 0.0 trans_lim: 0.0 type: GlobalRotScaleTrans
• point_cloud_range:
  • -54.0
  • -54.0
  • -5.0
  • 54.0
  • 54.0
  • 3.0 type: PointsRangeFilter
• mean:
  • 0.485
  • 0.456
  • 0.406 std:
  • 0.229
  • 0.224
  • 0.225 type: ImageNormalize
• classes:
  • car
  • truck
  • construction_vehicle
  • bus
  • trailer
  • barrier
  • motorcycle
  • bicycle
  • pedestrian
  • traffic_cone type: DefaultFormatBundle3D
• keys:
  • img
  • points
  • radar
  • gt_bboxes_3d
  • gt_labels_3d meta_keys:
  • camera_intrinsics
  • camera2ego
  • lidar2ego
  • lidar2camera
  • lidar2image
  • camera2lidar
  • img_aug_matrix
  • lidar_aug_matrix type: Collect3D
• keyframe_only: true type: GTDepth fp16: loss_scale: growth_interval: 2000 gt_paste_stop_epoch: -1 image_size:
  • 256
  • 704 input_modality: use_camera: true use_external: false use_lidar: true use_map: false use_radar: false load_augmented: null load_dim: 5 load_from: pretrained/lidar-only-det.pth log_config: hooks:
• type: TextLoggerHook
• type: TensorboardLoggerHook interval: 50 lr_config: min_lr_ratio: 0.001 policy: CosineAnnealing warmup: linear warmup_iters: 500 warmup_ratio: 0.33333333 map_classes:
  • drivable_area
  • ped_crossing
  • walkway
  • stop_line
  • carpark_area
  • divider max_epochs: 15 model: decoder: backbone: conv_cfg: bias: false type: Conv2d in_channels: 256 layer_nums:
    • 5
    • 5 layer_strides:
    • 1
    • 2 norm_cfg: eps: 0.001 momentum: 0.01 type: BN out_channels:
    • 128
    • 256 type: SECOND neck: in_channels:
    • 128
    • 256 norm_cfg: eps: 0.001 momentum: 0.01 type: BN out_channels:
    • 256
    • 256 type: SECONDFPN upsample_cfg: bias: false type: deconv upsample_strides:
    • 1
    • 2 use_conv_for_no_stride: true encoders: camera: backbone: attn_drop_rate: 0.0 convert_weights: true depths:
    • 2
    • 2
    • 6
    • 2 drop_path_rate: 0.2 drop_rate: 0.0 embed_dims: 96 init_cfg: checkpoint: pretrained/swint-nuimages-pretrained.pth type: Pretrained mlp_ratio: 4 num_heads:
    • 3
    • 6
    • 12
    • 24 out_indices:
    • 1
    • 2
    • 3 patch_norm: true qk_scale: null qkv_bias: true type: SwinTransformer window_size: 7 with_cp: false neck: act_cfg: inplace: true type: ReLU in_channels:
    • 192
    • 384
    • 768 norm_cfg: requires_grad: true type: BN2d num_outs: 3 out_channels: 256 start_level: 0 type: GeneralizedLSSFPN upsample_cfg: align_corners: false mode: bilinear vtransform: dbound:
    • 1.0
    • 60.0
    • 0.5 downsample: 2 feature_size:
    • 32
    • 88 image_size:
    • 256
    • 704 in_channels: 256 out_channels: 80 type: DepthLSSTransform xbound:
    • -54.0
    • 54.0
    • 0.3 ybound:
    • -54.0
    • 54.0
    • 0.3 zbound:
    • -10.0
    • 10.0
    • 20.0 lidar: backbone: block_type: basicblock encoder_channels:
    • • 16
      • 16
      • 32
    • • 32
      • 32
      • 64
    • • 64
      • 64
      • 128
    • • 128
      • 128 encoder_paddings:
    • • 0
      • 0
      • 1
    • • 0
      • 0
      • 1
    • • 0
      • 0
      • • 1
      • 1
      • 0
    • • 0
      • 0 in_channels: 5 order:
    • conv
    • norm
    • act output_channels: 128 sparse_shape:
    • 1440
    • 1440
    • 41 type: SparseEncoder voxelize: max_num_points: 10 max_voxels:
    • 120000
    • 160000 point_cloud_range:
    • -54.0
    • -54.0
    • -5.0
    • 54.0
    • 54.0
    • 3.0 voxel_size:
    • 0.075
    • 0.075
    • 0.2 fuser: in_channels:
  • 80
  • 256 out_channels: 256 type: ConvFuser heads: map: null object: activation: relu auxiliary: true bbox_coder: code_size: 10 out_size_factor: 8 pc_range:
    • -54.0
    • -54.0 post_center_range:
    • -61.2
    • -61.2
    • -10.0
    • 61.2
    • 61.2
    • 10.0 score_threshold: 0.0 type: TransFusionBBoxCoder voxel_size:
    • 0.075
    • 0.075 bn_momentum: 0.1 common_heads: center:
    • 2
    • 2 dim:
    • 3
    • 2 height:
    • 1
    • 2 rot:
    • 2
    • 2 vel:
    • 2
    • 2 dropout: 0.1 ffn_channel: 256 hidden_channel: 128 in_channels: 512 loss_bbox: loss_weight: 0.25 reduction: mean type: L1Loss loss_cls: alpha: 0.25 gamma: 2.0 loss_weight: 1.0 reduction: mean type: FocalLoss use_sigmoid: true loss_heatmap: loss_weight: 1.0 reduction: mean type: GaussianFocalLoss nms_kernel_size: 3 num_classes: 10 num_decoder_layers: 1 num_heads: 8 num_proposals: 200 test_cfg: dataset: nuScenes grid_size:
    • 1440
    • 1440
    • 41 nms_type: null out_size_factor: 8 pc_range:
    • -54.0
    • -54.0 voxel_size:
    • 0.075
    • 0.075 train_cfg: assigner: cls_cost: alpha: 0.25 gamma: 2.0 type: FocalLossCost weight: 0.15 iou_calculator: coordinate: lidar type: BboxOverlaps3D iou_cost: type: IoU3DCost weight: 0.25 reg_cost: type: BBoxBEVL1Cost weight: 0.25 type: HungarianAssigner3D code_weights:
    • 1.0
    • 1.0
    • 1.0
    • 1.0
    • 1.0
    • 1.0
    • 1.0
    • 1.0
    • 0.2
    • 0.2 dataset: nuScenes gaussian_overlap: 0.1 grid_size:
    • 1440
    • 1440
    • 41 min_radius: 2 out_size_factor: 8 point_cloud_range:
    • -54.0
    • -54.0
    • -5.0
    • 54.0
    • 54.0
    • 3.0 pos_weight: -1 voxel_size:
    • 0.075
    • 0.075
    • 0.2 type: TransFusionHead type: BEVFusion momentum_config: policy: cyclic object_classes:
  • car
  • truck
  • construction_vehicle
  • bus
  • trailer
  • barrier
  • motorcycle
  • bicycle
  • pedestrian
  • traffic_cone optimizer: lr: 0.0002 type: AdamW weight_decay: 0.01 optimizer_config: grad_clip: max_norm: 35 norm_type: 2 point_cloud_range:
  • -54.0
  • -54.0
  • -5.0
  • 54.0
  • 54.0
  • 3.0 radar_compensate_velocity: true radar_filtering: none radar_jitter: 0 radar_max_points: 2500 radar_normalize: false radar_sweeps: 6 radar_use_dims:
  • 0
  • 1
  • 2
  • 5
  • 8
  • 9
  • 18
  • 19
  • 20
  • 21
  • 22
  • 23
  • 24
  • 25
  • 26
  • 27
  • 28
  • 29
  • 30
  • 31
  • 32
  • 33
  • 34
  • 35
  • 36
  • 37
  • 38
  • 39
  • 40
  • 41
  • 42
  • 43
  • 44
  • 45
  • 46
  • 47
  • 48
  • 49
  • 50
  • 51
  • 52
  • 53
  • 54
  • 55
  • 56 radar_voxel_size:
  • 0.8
  • 0.8
  • 8 reduce_beams: 32 resume_from: null run_dir: runs/run-7423be99 runner: max_epochs: 15 type: CustomEpochBasedRunner seed: 0 test_pipeline:
  • to_float32: true type: LoadMultiViewImageFromFiles
  • coord_type: LIDAR load_augmented: null load_dim: 5 reduce_beams: 32 type: LoadPointsFromFile use_dim: 5
  • load_augmented: null load_dim: 5 pad_empty_sweeps: true reduce_beams: 32 remove_close: true sweeps_num: 9 type: LoadPointsFromMultiSweeps use_dim: 5
  • compensate_velocity: true filtering: none load_dim: 18 max_num: 2500 normalize: false sweeps_num: 6 type: LoadRadarPointsMultiSweeps use_dim:
• 0
• 1
• 2
• 5
• 8
• 9
• 18
• 19
• 20
• 21
• 22
• 23
• 24
• 25
• 26
• 27
• 28
• 29
• 30
• 31
• 32
• 33
• 34
• 35
• 36
• 37
• 38
• 39
• 40
• 41
• 42
• 43
• 44
• 45
• 46
• 47
• 48
• 49
• 50
• 51
• 52
• 53
• 54
• 55
• 56
  • type: LoadAnnotations3D with_attr_label: false with_bbox_3d: true with_label_3d: true
  • bot_pct_lim:
• 0.0
• 0.0 final_dim:
• 256
• 704 is_train: false rand_flip: false resize_lim:
• 0.48
• 0.48 rot_lim:
• 0.0
• 0.0 type: ImageAug3D
  • is_train: false resize_lim:
• 1.0
• 1.0 rot_lim:
• 0.0
• 0.0 trans_lim: 0.0 type: GlobalRotScaleTrans
  • point_cloud_range:
• -54.0
• -54.0
• -5.0
• 54.0
• 54.0
• 3.0 type: PointsRangeFilter
  • mean:
• 0.485
• 0.456
• 0.406 std:
• 0.229
• 0.224
• 0.225 type: ImageNormalize
  • classes:
• car
• truck
• construction_vehicle
• bus
• trailer
• barrier
• motorcycle
• bicycle
• pedestrian
• traffic_cone type: DefaultFormatBundle3D
  • keys:
• img
• points
• radar
• gt_bboxes_3d
• gt_labels_3d meta_keys:
• camera_intrinsics
• camera2ego
• lidar2ego
• lidar2camera
• lidar2image
• camera2lidar
• img_aug_matrix
• lidar_aug_matrix type: Collect3D
  • keyframe_only: true type: GTDepth train_pipeline:
  • to_float32: true type: LoadMultiViewImageFromFiles
  • coord_type: LIDAR load_augmented: null load_dim: 5 reduce_beams: 32 type: LoadPointsFromFile use_dim: 5
  • load_augmented: null load_dim: 5 pad_empty_sweeps: true reduce_beams: 32 remove_close: true sweeps_num: 9 type: LoadPointsFromMultiSweeps use_dim: 5
  • compensate_velocity: true filtering: none load_dim: 18 max_num: 2500 normalize: false sweeps_num: 6 type: LoadRadarPointsMultiSweeps use_dim:
• 0
• 1
• 2
• 5
• 8
• 9
• 18
• 19
• 20
• 21
• 22
• 23
• 24
• 25
• 26
• 27
• 28
• 29
• 30
• 31
• 32
• 33
• 34
• 35
• 36
• 37
• 38
• 39
• 40
• 41
• 42
• 43
• 44
• 45
• 46
• 47
• 48
• 49
• 50
• 51
• 52
• 53
• 54
• 55
• 56
  • type: LoadAnnotations3D with_attr_label: false with_bbox_3d: true with_label_3d: true
  • db_sampler: classes:
  • car
  • truck
  • construction_vehicle
  • bus
  • trailer
  • barrier
  • motorcycle
  • bicycle
  • pedestrian
  • traffic_cone dataset_root: data/nuscenes/ info_path: data/nuscenes/nuscenes_dbinfos_train.pkl points_loader: coord_type: LIDAR load_dim: 5 reduce_beams: 32 type: LoadPointsFromFile use_dim: 5 prepare: filter_by_difficulty:
    • -1 filter_by_min_points: barrier: 5 bicycle: 5 bus: 5 car: 5 construction_vehicle: 5 motorcycle: 5 pedestrian: 5 traffic_cone: 5 trailer: 5 truck: 5 rate: 1.0 sample_groups: barrier: 2 bicycle: 6 bus: 4 car: 2 construction_vehicle: 7 motorcycle: 6 pedestrian: 2 traffic_cone: 2 trailer: 6 truck: 3 stop_epoch: -1 type: ObjectPaste
  • bot_pct_lim:
• 0.0
• 0.0 final_dim:
• 256
• 704 is_train: true rand_flip: true resize_lim:
• 0.38
• 0.55 rot_lim:
• -5.4
• 5.4 type: ImageAug3D
  • is_train: true resize_lim:
• 0.9
• 1.1 rot_lim:
• -0.78539816
• 0.78539816 trans_lim: 0.5 type: GlobalRotScaleTrans
  • type: RandomFlip3D
  • point_cloud_range:
• -54.0
• -54.0
• -5.0
• 54.0
• 54.0
• 3.0 type: PointsRangeFilter
  • point_cloud_range:
• -54.0
• -54.0
• -5.0
• 54.0
• 54.0
• 3.0 type: ObjectRangeFilter
  • classes:
• car
• truck
• construction_vehicle
• bus
• trailer
• barrier
• motorcycle
• bicycle
• pedestrian
• traffic_cone type: ObjectNameFilter
  • mean:
• 0.485
• 0.456
• 0.406 std:
• 0.229
• 0.224
• 0.225 type: ImageNormalize
  • fixed_prob: true max_epoch: 15 mode: 1 offset: false prob: 0.0 ratio: 0.5 rotate: 1 type: GridMask use_h: true use_w: true
  • type: PointShuffle
  • classes:
• car
• truck
• construction_vehicle
• bus
• trailer
• barrier
• motorcycle
• bicycle
• pedestrian
• traffic_cone type: DefaultFormatBundle3D
  • keys:
• img
• points
• radar
• gt_bboxes_3d
• gt_labels_3d meta_keys:
• camera_intrinsics
• camera2ego
• lidar2ego
• lidar2camera
• lidar2image
• camera2lidar
• img_aug_matrix
• lidar_aug_matrix type: Collect3D
  • keyframe_only: true type: GTDepth use_dim: 5 voxel_size:
  • 0.075
  • 0.075
  • 0.2

[Tamako668]

Hello, have you solved this problem? I encountered the same problem.

[JohnFuguiWang]

Hello, have you solved this problem? I encountered the same problem.

Honestly speaking, I still don't know the reason of this problem now. I reinstalled the conda environment following this blog , and ran the model again using nuscenes-mini. The result map was around 0.47 and stayed stable. I think maybe this result is normal. You can try to reinstall your conda environment too and run the model again. If it's possible you can notify me of your new result, and we can check if our results are normal. Thanks in advance.