[Fix] Fix ImVoxelNet and KittiMetric & refactor configs of ImVoxelNet

configs/imvoxelnet/imvoxelnet_8xb4_kitti-3d-car.py

@@ -1,3 +1,7 @@
+_base_ = [
+ '../_base_/schedules/mmdet-schedule-1x.py', '../_base_/default_runtime.py'
+]
+
 model = dict(
  type='ImVoxelNet',
  data_preprocessor=dict(
@@ -151,7 +155,8 @@
 # optimizer
 optim_wrapper = dict(
  type='OptimWrapper',
- optimizer=dict(type='AdamW', lr=0.0001, weight_decay=0.0001),
+ optimizer=dict(
+ _delete_=True, type='AdamW', lr=0.0001, weight_decay=0.0001),
  paramwise_cfg=dict(
  custom_keys={'backbone': dict(lr_mult=0.1, decay_mult=1.0)}),
  clip_grad=dict(max_norm=35., norm_type=2))
@@ -166,30 +171,7 @@
 ]
 
 # hooks
-default_hooks = dict(
- timer=dict(type='IterTimerHook'),
- logger=dict(type='LoggerHook', interval=50),
- param_scheduler=dict(type='ParamSchedulerHook'),
- checkpoint=dict(type='CheckpointHook', interval=1, max_keep_ckpts=1),
- sampler_seed=dict(type='DistSamplerSeedHook'),
-)
-
-# training schedule for 2x
-train_cfg = dict(type='EpochBasedTrainLoop', max_epochs=12, val_interval=1)
-val_cfg = dict(type='ValLoop')
-test_cfg = dict(type='TestLoop')
+default_hooks = dict(checkpoint=dict(type='CheckpointHook', max_keep_ckpts=1))
 
 # runtime
-default_scope = 'mmdet3d'
-
-env_cfg = dict(
- cudnn_benchmark=False,
- mp_cfg=dict(mp_start_method='fork', opencv_num_threads=0),
- dist_cfg=dict(backend='nccl'),
-)
-
-log_level = 'INFO'
-load_from = None
-resume = False
-dist_params = dict(backend='nccl')
 find_unused_parameters = True # only 1 of 4 FPN outputs is used

mmdet3d/evaluation/metrics/kitti_metric.py

@@ -575,7 +575,7 @@ def convert_valid_bboxes(self, box_dict: dict, info: dict):
  box_preds = box_dict['bboxes_3d']
  scores = box_dict['scores_3d']
  labels = box_dict['labels_3d']
- sample_idx = info['sample_id']
+ sample_idx = info['sample_idx']
  box_preds.limit_yaw(offset=0.5, period=np.pi * 2)
 
  if len(box_preds) == 0:

mmdet3d/models/detectors/base.py

@@ -89,7 +89,7 @@ def forward(self,
  raise RuntimeError(f'Invalid mode "{mode}". '
  'Only supports loss, predict and tensor mode')
 
- def convert_to_datasample(
+ def add_pred_to_datasample(
  self,
  data_samples: SampleList,
  data_instances_3d: OptInstanceList = None,

mmdet3d/models/detectors/fcos_mono3d.py

@@ -95,6 +95,7 @@ def predict(self,
  x = self.extract_feat(batch_inputs_dict)
  results_list, results_list_2d = self.bbox_head.predict(
  x, batch_data_samples, rescale=rescale)
- predictions = self.convert_to_datasample(batch_data_samples,
- results_list, results_list_2d)
+ predictions = self.add_pred_to_datasample(batch_data_samples,
+ results_list,
+ results_list_2d)
  return predictions

mmdet3d/models/detectors/groupfree3dnet.py

@@ -82,6 +82,6 @@ def predict(self, batch_inputs_dict: dict, batch_data_samples: SampleList,
  points = batch_inputs_dict['points']
  results_list = self.bbox_head.predict(points, x, batch_data_samples,
  **kwargs)
- predictions = self.convert_to_datasample(batch_data_samples,
- results_list)
+ predictions = self.add_pred_to_datasample(batch_data_samples,
+  results_list)
  return predictions

mmdet3d/models/detectors/h3dnet.py

@@ -154,4 +154,4 @@ def predict(
  feats_dict,
  batch_data_samples,
  suffix='_optimized')
- return self.convert_to_datasample(batch_data_samples, results_list)
+ return self.add_pred_to_datasample(batch_data_samples, results_list)

mmdet3d/models/detectors/imvotenet.py

@@ -433,7 +433,7 @@ def predict(self, batch_inputs_dict: Dict[str, Optional[Tensor]],
  if points is None:
  assert imgs is not None
  results_2d = self.predict_img_only(imgs, batch_data_samples)
- return self.convert_to_datasample(
+ return self.add_pred_to_datasample(
  batch_data_samples, data_instances_2d=results_2d)
 
  else:
@@ -488,7 +488,7 @@ def predict(self, batch_inputs_dict: Dict[str, Optional[Tensor]],
  batch_data_samples,
  rescale=True)
 
- return self.convert_to_datasample(batch_data_samples, results_3d)
+ return self.add_pred_to_datasample(batch_data_samples, results_3d)
 
  def predict_img_only(self,
  imgs: Tensor,

mmdet3d/models/detectors/imvoxelnet.py

@@ -3,15 +3,15 @@
 
 import torch
 
+from mmdet3d.models.detectors import Base3DDetector
 from mmdet3d.models.layers.fusion_layers.point_fusion import point_sample
 from mmdet3d.registry import MODELS, TASK_UTILS
 from mmdet3d.structures.det3d_data_sample import SampleList
-from mmdet3d.utils import ConfigType, InstanceList, OptConfigType
-from mmdet.models.detectors import BaseDetector
+from mmdet3d.utils import ConfigType, OptConfigType
 
 
 @MODELS.register_module()
-class ImVoxelNet(BaseDetector):
+class ImVoxelNet(Base3DDetector):
  r"""`ImVoxelNet <https://arxiv.org/abs/2106.01178>`_.
 
  Args:
@@ -57,31 +57,6 @@ def __init__(self,
  self.train_cfg = train_cfg
  self.test_cfg = test_cfg
 
- def convert_to_datasample(self, data_samples: SampleList,
- data_instances: InstanceList) -> SampleList:
- """ Convert results list to `Det3DDataSample`.
- Args:
- inputs (list[:obj:`Det3DDataSample`]): The input data.
- data_instances (list[:obj:`InstanceData`]): 3D Detection
- results of each image.
- Returns:
- list[:obj:`Det3DDataSample`]: 3D Detection results of the
- input images. Each Det3DDataSample usually contain
- 'pred_instances_3d'. And the ``pred_instances_3d`` usually
- contains following keys.
-
- - scores_3d (Tensor): Classification scores, has a shape
- (num_instance, )
- - labels_3d (Tensor): Labels of bboxes, has a shape
- (num_instances, ).
- - bboxes_3d (Tensor): Contains a tensor with shape
- (num_instances, C) where C >=7.
- """
- for data_sample, pred_instances_3d in zip(data_samples,
- data_instances):
- data_sample.pred_instances_3d = pred_instances_3d
- return data_samples
-
  def extract_feat(self, batch_inputs_dict: dict,
  batch_data_samples: SampleList):
  """Extract 3d features from the backbone -> fpn -> 3d projection.
@@ -185,8 +160,8 @@ def predict(self, batch_inputs_dict: dict, batch_data_samples: SampleList,
  """
  x = self.extract_feat(batch_inputs_dict, batch_data_samples)
  results_list = self.bbox_head.predict(x, batch_data_samples, **kwargs)
- predictions = self.convert_to_datasample(batch_data_samples,
- results_list)
+ predictions = self.add_pred_to_datasample(batch_data_samples,
+  results_list)
  return predictions
 
  def _forward(self, batch_inputs_dict: dict, batch_data_samples: SampleList,

mmdet3d/models/detectors/mvx_two_stage.py

@@ -401,7 +401,7 @@ def predict(self, batch_inputs_dict: Dict[str, Optional[Tensor]],
  else:
  results_list_2d = None
 
- detsamples = self.convert_to_datasample(batch_data_samples,
- results_list_3d,
- results_list_2d)
+ detsamples = self.add_pred_to_datasample(batch_data_samples,
+  results_list_3d,
+  results_list_2d)
  return detsamples

mmdet3d/models/detectors/single_stage.py

@@ -108,8 +108,8 @@ def predict(self, batch_inputs_dict: dict, batch_data_samples: SampleList,
  """
  x = self.extract_feat(batch_inputs_dict)
  results_list = self.bbox_head.predict(x, batch_data_samples, **kwargs)
- predictions = self.convert_to_datasample(batch_data_samples,
- results_list)
+ predictions = self.add_pred_to_datasample(batch_data_samples,
+  results_list)
  return predictions
 
  def _forward(self,

mmdet3d/models/detectors/single_stage_mono3d.py

@@ -18,7 +18,7 @@ class SingleStageMono3DDetector(SingleStageDetector):
  boxes on the output features of the backbone+neck.
  """
 
- def convert_to_datasample(
+ def add_pred_to_datasample(
  self,
  data_samples: SampleList,
  data_instances_3d: OptInstanceList = None,

mmdet3d/models/detectors/two_stage.py

@@ -161,8 +161,8 @@ def predict(self, batch_inputs_dict: dict, batch_data_samples: SampleList,
  batch_data_samples)
 
  # connvert to Det3DDataSample
- results_list = self.convert_to_datasample(batch_data_samples,
- results_list)
+ results_list = self.add_pred_to_datasample(batch_data_samples,
+  results_list)
 
  return results_list
 

mmdet3d/models/detectors/votenet.py

@@ -99,8 +99,8 @@ def predict(self, batch_inputs_dict: Dict[str, Optional[Tensor]],
  points = batch_inputs_dict['points']
  results_list = self.bbox_head.predict(points, feats_dict,
  batch_data_samples, **kwargs)
- data_3d_samples = self.convert_to_datasample(batch_data_samples,
- results_list)
+ data_3d_samples = self.add_pred_to_datasample(batch_data_samples,
+  results_list)
  return data_3d_samples
 
  def aug_test(self, aug_inputs_list: List[dict],
@@ -143,6 +143,6 @@ def aug_test(self, aug_inputs_list: List[dict],
  self.bbox_head.test_cfg)
 
  merged_results = InstanceData(**merged_results_dict)
- data_3d_samples = self.convert_to_datasample(batch_data_samples,
- [merged_results])
+ data_3d_samples = self.add_pred_to_datasample(batch_data_samples,
+  [merged_results])
  return data_3d_samples