[Refactor]: remove legacy builder.py and update corresponding unit tests (#2161)

* fix circle ci

* delete model builder

* fix comments

tests/test_models/test_losses/test_uncertain_smooth_l1_loss.py

@@ -2,7 +2,7 @@
 import pytest
 import torch
 
-from mmdet3d.models.builder import build_loss
+from mmdet3d.registry import MODELS
 
 
 def test_uncertain_smooth_l1_loss():
@@ -21,7 +21,7 @@ def test_uncertain_smooth_l1_loss():
     # test uncertain l1 loss
     uncertain_l1_loss_cfg = dict(
         type='UncertainL1Loss', alpha=1.0, reduction='mean', loss_weight=1.0)
-    uncertain_l1_loss = build_loss(uncertain_l1_loss_cfg)
+    uncertain_l1_loss = MODELS.build(uncertain_l1_loss_cfg)
     mean_l1_loss = uncertain_l1_loss(pred, target, sigma)
     expected_l1_loss = torch.tensor(4.7069)
     assert torch.allclose(mean_l1_loss, expected_l1_loss, atol=1e-4)
@@ -33,7 +33,7 @@ def test_uncertain_smooth_l1_loss():
         beta=0.5,
         reduction='mean',
         loss_weight=1.0)
-    uncertain_smooth_l1_loss = build_loss(uncertain_smooth_l1_loss_cfg)
+    uncertain_smooth_l1_loss = MODELS.build(uncertain_smooth_l1_loss_cfg)
     mean_smooth_l1_loss = uncertain_smooth_l1_loss(pred, target, sigma)
     expected_smooth_l1_loss = torch.tensor(3.9795)
     assert torch.allclose(

tests/test_models/test_middle_encoders/test_sparse_encoders.py

@@ -2,7 +2,7 @@
 import pytest
 import torch
 
-from mmdet3d.models.builder import build_middle_encoder
+from mmdet3d.registry import MODELS
 
 
 def test_sparse_encoder():
@@ -19,7 +19,7 @@ def test_sparse_encoder():
                                                                        1)),
         block_type='basicblock')
 
-    sparse_encoder = build_middle_encoder(sparse_encoder_cfg).cuda()
+    sparse_encoder = MODELS.build(sparse_encoder_cfg).cuda()
     voxel_features = torch.rand([207842, 5]).cuda()
     coors = torch.randint(0, 4, [207842, 4]).cuda()
 
@@ -41,7 +41,7 @@ def test_sparse_encoder_for_ssd():
                                                                        1)),
         block_type='basicblock')
 
-    sparse_encoder = build_middle_encoder(sparse_encoder_for_ssd_cfg).cuda()
+    sparse_encoder = MODELS.build(sparse_encoder_for_ssd_cfg).cuda()
     voxel_features = torch.rand([207842, 5]).cuda()
     coors = torch.randint(0, 4, [207842, 4]).cuda()
 

tests/test_models/test_necks/test_dla_neck.py

@@ -2,7 +2,7 @@
 
 import torch
 
-from mmdet3d.models.builder import build_neck
+from mmdet3d.registry import MODELS
 
 
 def test_dla_neck():
@@ -19,7 +19,7 @@ def test_dla_neck():
             start_level=2,
             end_level=5,
             norm_cfg=dict(type='GN', num_groups=32))
-        neck = build_neck(neck_cfg)
+        neck = MODELS.build(neck_cfg)
         neck.init_weights()
         neck.cuda()
         feats = [
@@ -37,7 +37,7 @@ def test_dla_neck():
             end_level=5,
             norm_cfg=dict(type='GN', num_groups=32),
             use_dcn=False)
-        neck = build_neck(neck_cfg)
+        neck = MODELS.build(neck_cfg)
         neck.init_weights()
         feats = [
             torch.rand(4, in_channels[i], feat_sizes[i], feat_sizes[i])

tests/test_models/test_necks/test_imvoxel_neck.py

 import pytest
 import torch
 
-from mmdet3d.models.builder import build_neck
+from mmdet3d.registry import MODELS
 
 
 def test_imvoxel_neck():
@@ -10,7 +10,7 @@ def test_imvoxel_neck():
 
     neck_cfg = dict(
         type='OutdoorImVoxelNeck', in_channels=64, out_channels=256)
-    neck = build_neck(neck_cfg).cuda()
+    neck = MODELS.build(neck_cfg).cuda()
     inputs = torch.rand([1, 64, 216, 248, 12], device='cuda')
     outputs = neck(inputs)
     assert outputs[0].shape == (1, 256, 248, 216)

tests/test_models/test_necks/test_pointnet2_fp_neck.py

 import pytest
 import torch
 
-from mmdet3d.models.builder import build_neck
+from mmdet3d.registry import MODELS
 
 
 def test_pointnet2_fp_neck():
@@ -22,7 +22,7 @@ def test_pointnet2_fp_neck():
         fp_channels=((1536, 512, 512), (768, 512, 512), (608, 256, 256),
                      (257, 128, 128)))
 
-    neck = build_neck(neck_cfg)
+    neck = MODELS.build(neck_cfg)
     neck.init_weights()
 
     if torch.cuda.is_available():

tests/test_models/test_necks/test_second_fpn.py

@@ -2,7 +2,7 @@
 import pytest
 import torch
 
-from mmdet3d.models.builder import build_backbone, build_neck
+from mmdet3d.registry import MODELS
 
 
 def test_secfpn():
@@ -12,8 +12,7 @@ def test_secfpn():
         upsample_strides=[1, 2],
         out_channels=[4, 6],
     )
-    from mmdet3d.models.builder import build_neck
-    neck = build_neck(neck_cfg)
+    neck = MODELS.build(neck_cfg)
     assert neck.deblocks[0][0].in_channels == 2
     assert neck.deblocks[1][0].in_channels == 3
     assert neck.deblocks[0][0].out_channels == 4
@@ -29,7 +28,7 @@ def test_secfpn():
         out_channels=[2, 2],
     )
     with pytest.raises(AssertionError):
-        build_neck(neck_cfg)
+        MODELS.build(neck_cfg)
 
     neck_cfg = dict(
         type='SECONDFPN',
@@ -38,7 +37,7 @@ def test_secfpn():
         out_channels=[2, 2],
     )
     with pytest.raises(AssertionError):
-        build_neck(neck_cfg)
+        MODELS.build(neck_cfg)
 
 
 def test_centerpoint_fpn():
@@ -52,7 +51,7 @@ def test_centerpoint_fpn():
         norm_cfg=dict(type='BN', eps=1e-3, momentum=0.01),
         conv_cfg=dict(type='Conv2d', bias=False))
 
-    second = build_backbone(second_cfg)
+    second = MODELS.build(second_cfg)
 
     # centerpoint usage of fpn
     centerpoint_fpn_cfg = dict(
@@ -71,9 +70,9 @@ def test_centerpoint_fpn():
         upsample_strides=[1, 2, 4],
         out_channels=[2, 2, 2])
 
-    second_fpn = build_neck(fpn_cfg)
+    second_fpn = MODELS.build(fpn_cfg)
 
-    centerpoint_second_fpn = build_neck(centerpoint_fpn_cfg)
+    centerpoint_second_fpn = MODELS.build(centerpoint_fpn_cfg)
 
     input = torch.rand([2, 2, 32, 32])
     sec_output = second(input)

tests/test_models/test_voxel_encoders/test_pillar_encoder.py

@@ -2,7 +2,7 @@
 import pytest
 import torch
 
-from mmdet3d.models.builder import build_voxel_encoder
+from mmdet3d.registry import MODELS
 
 
 def test_pillar_feature_net():
@@ -16,7 +16,7 @@ def test_pillar_feature_net():
         voxel_size=(0.2, 0.2, 8),
         point_cloud_range=(-51.2, -51.2, -5.0, 51.2, 51.2, 3.0),
         norm_cfg=dict(type='BN1d', eps=1e-3, momentum=0.01))
-    pillar_feature_net = build_voxel_encoder(pillar_feature_net_cfg)
+    pillar_feature_net = MODELS.build(pillar_feature_net_cfg)
 
     features = torch.rand([97297, 20, 5])
     num_voxels = torch.randint(1, 100, [97297])

tests/test_models/test_voxel_encoders/test_voxel_encoders.py

@@ -2,14 +2,14 @@
 import pytest
 import torch
 
-from mmdet3d.models.builder import build_voxel_encoder
+from mmdet3d.registry import MODELS
 
 
 def test_hard_simple_VFE():
     if not torch.cuda.is_available():
         pytest.skip('test requires GPU and torch+cuda')
     hard_simple_VFE_cfg = dict(type='HardSimpleVFE', num_features=5)
-    hard_simple_VFE = build_voxel_encoder(hard_simple_VFE_cfg)
+    hard_simple_VFE = MODELS.build(hard_simple_VFE_cfg)
     features = torch.rand([240000, 10, 5])
     num_voxels = torch.randint(1, 10, [240000])
 

tools/analysis_tools/benchmark.py

@@ -7,8 +7,7 @@ from mmcv import Config
 from mmcv.parallel import MMDataParallel
 from mmengine.runner import load_checkpoint
 
-from mmdet3d.datasets import build_dataset
-from mmdet3d.models import build_detector
+from mmdet3d.registry import DATASETS, MODELS
 from tools.misc.fuse_conv_bn import fuse_module
 
 
@@ -40,7 +39,7 @@ def main():
 
     # build the dataloader
     # TODO: support multiple images per gpu (only minor changes are needed)
-    dataset = build_dataset(cfg.data.test)
+    dataset = DATASETS.build(cfg.data.test)
 
     # TODO fix this
     def build_dataloader():
@@ -55,7 +54,7 @@ def main():
 
     # build the model and load checkpoint
     cfg.model.train_cfg = None
-    model = build_detector(cfg.model, test_cfg=cfg.get('test_cfg'))
+    model = MODELS.build(cfg.model, test_cfg=cfg.get('test_cfg'))
     load_checkpoint(model, args.checkpoint, map_location='cpu')
     if args.fuse_conv_bn:
         model = fuse_module(model)

tools/dataset_converters/create_gt_database.py

@@ -11,7 +11,7 @@ from mmengine import track_iter_progress
 from pycocotools import mask as maskUtils
 from pycocotools.coco import COCO
 
-from mmdet3d.datasets import build_dataset
+from mmdet3d.registry import DATASETS
 from mmdet3d.structures.ops import box_np_ops as box_np_ops
 
 
@@ -218,7 +218,7 @@ def create_groundtruth_database(dataset_class_name,
                     file_client_args=file_client_args)
             ])
 
-    dataset = build_dataset(dataset_cfg)
+    dataset = DATASETS.build(dataset_cfg)
 
     if database_save_path is None:
         database_save_path = osp.join(data_path, f'{info_prefix}_gt_database')
@@ -587,7 +587,7 @@ class GTDatabaseCreater:
                         file_client_args=file_client_args)
                 ])
 
-        self.dataset = build_dataset(dataset_cfg)
+        self.dataset = DATASETS.build(dataset_cfg)
         self.pipeline = self.dataset.pipeline
         if self.database_save_path is None:
             self.database_save_path = osp.join(

tools/misc/visualize_results.py

@@ -2,9 +2,9 @@
 import argparse
 
 import mmengine
-from mmcv import Config
+from mmengine import Config
 
-from mmdet3d.datasets import build_dataset
+from mmdet3d.registry import DATASETS
 
 
 def parse_args():
@@ -30,7 +30,7 @@ def main():
     cfg.data.test.test_mode = True
 
     # build the dataset
-    dataset = build_dataset(cfg.data.test)
+    dataset = DATASETS.build(cfg.data.test)
     results = mmengine.load(args.result)
 
     if getattr(dataset, 'show', None) is not None:

tools/model_converters/convert_h3dnet_checkpoints.py

@@ -6,7 +6,7 @@ import torch
 from mmcv import Config
 from mmengine.runner import load_state_dict
 
-from mmdet3d.models import build_detector
+from mmdet3d.registry import MODELS
 
 
 def parse_args():
@@ -103,7 +103,7 @@ def main():
     checkpoint = torch.load(args.checkpoint)
     cfg = parse_config(checkpoint['meta']['config'])
     # Build the model and load checkpoint
-    model = build_detector(
+    model = MODELS.build(
         cfg.model,
         train_cfg=cfg.get('train_cfg'),
         test_cfg=cfg.get('test_cfg'))

tools/model_converters/convert_votenet_checkpoints.py

@@ -6,7 +6,7 @@ import torch
 from mmengine import Config
 from mmengine.runner import load_state_dict
 
-from mmdet3d.models import build_detector
+from mmdet3d.registry import MODELS
 
 
 def parse_args():
@@ -79,7 +79,7 @@ def main():
     checkpoint = torch.load(args.checkpoint)
     cfg = parse_config(checkpoint['meta']['config'])
     # Build the model and load checkpoint
-    model = build_detector(
+    model = MODELS.build(
         cfg.model,
         train_cfg=cfg.get('train_cfg'),
         test_cfg=cfg.get('test_cfg'))