Bump version to v1.1.0rc2

Bump to v1.1.0rc2

tests/test_models/test_detectors/test_imvotenet.py

@@ -4,8 +4,8 @@ import torch
 from mmengine import DefaultScope
 
 from mmdet3d.registry import MODELS
-from tests.utils.model_utils import (_create_detector_inputs,
-                                     _get_detector_cfg, _setup_seed)
+from mmdet3d.testing import (create_detector_inputs, get_detector_cfg,
+                             setup_seed)
 
 
 class TestImvoteNet(unittest.TestCase):
@@ -15,12 +15,12 @@ class TestImvoteNet(unittest.TestCase):
 
         assert hasattr(mmdet3d.models, 'ImVoteNet')
         DefaultScope.get_instance('test_imvotenet_img', scope_name='mmdet3d')
-        _setup_seed(0)
-        votenet_net_cfg = _get_detector_cfg(
+        setup_seed(0)
+        votenet_net_cfg = get_detector_cfg(
             'imvotenet/imvotenet_faster-rcnn-r50_fpn_4xb2_sunrgbd-3d.py')
         model = MODELS.build(votenet_net_cfg)
 
-        packed_inputs = _create_detector_inputs(
+        packed_inputs = create_detector_inputs(
             with_points=False, with_img=True, img_size=128)
 
         if torch.cuda.is_available():
@@ -49,12 +49,12 @@ class TestImvoteNet(unittest.TestCase):
 
         assert hasattr(mmdet3d.models, 'ImVoteNet')
         DefaultScope.get_instance('test_imvotenet', scope_name='mmdet3d')
-        _setup_seed(0)
-        votenet_net_cfg = _get_detector_cfg(
+        setup_seed(0)
+        votenet_net_cfg = get_detector_cfg(
             'imvotenet/imvotenet_stage2_8xb16_sunrgbd-3d.py')
         model = MODELS.build(votenet_net_cfg)
 
-        packed_inputs = _create_detector_inputs(
+        packed_inputs = create_detector_inputs(
             with_points=True,
             with_img=True,
             img_size=128,

tests/test_models/test_detectors/test_imvoxelnet.py

@@ -4,8 +4,8 @@ import torch
 from mmengine import DefaultScope
 
 from mmdet3d.registry import MODELS
-from tests.utils.model_utils import (_create_detector_inputs,
-                                     _get_detector_cfg, _setup_seed)
+from mmdet3d.testing import (create_detector_inputs, get_detector_cfg,
+                             setup_seed)
 
 
 class TestImVoxelNet(unittest.TestCase):
@@ -15,12 +15,12 @@ class TestImVoxelNet(unittest.TestCase):
 
         assert hasattr(mmdet3d.models, 'ImVoxelNet')
         DefaultScope.get_instance('test_ImVoxelNet', scope_name='mmdet3d')
-        _setup_seed(0)
-        imvoxel_net_cfg = _get_detector_cfg(
+        setup_seed(0)
+        imvoxel_net_cfg = get_detector_cfg(
             'imvoxelnet/imvoxelnet_8xb4_kitti-3d-car.py')
         model = MODELS.build(imvoxel_net_cfg)
         num_gt_instance = 1
-        packed_inputs = _create_detector_inputs(
+        packed_inputs = create_detector_inputs(
             with_points=False,
             with_img=True,
             img_size=(128, 128),

tests/test_models/test_detectors/test_mvxnet.py

@@ -4,8 +4,8 @@ import torch
 from mmengine import DefaultScope
 
 from mmdet3d.registry import MODELS
-from tests.utils.model_utils import (_create_detector_inputs,
-                                     _get_detector_cfg, _setup_seed)
+from mmdet3d.testing import (create_detector_inputs, get_detector_cfg,
+                             setup_seed)
 
 
 class TestMVXNet(unittest.TestCase):
@@ -15,14 +15,14 @@ class TestMVXNet(unittest.TestCase):
 
         assert hasattr(mmdet3d.models, 'DynamicMVXFasterRCNN')
 
-        _setup_seed(0)
+        setup_seed(0)
         DefaultScope.get_instance('test_mvxnet', scope_name='mmdet3d')
-        mvx_net_cfg = _get_detector_cfg(
+        mvx_net_cfg = get_detector_cfg(
             'mvxnet/mvxnet_fpn_dv_second_secfpn_8xb2-80e_kitti-3d-3class.py'  # noqa
         )
         model = MODELS.build(mvx_net_cfg)
         num_gt_instance = 1
-        packed_inputs = _create_detector_inputs(
+        packed_inputs = create_detector_inputs(
             with_img=False, num_gt_instance=num_gt_instance, points_feat_dim=4)
 
         if torch.cuda.is_available():

tests/test_models/test_detectors/test_parta2.py

@@ -4,8 +4,8 @@ import torch
 from mmengine import DefaultScope
 
 from mmdet3d.registry import MODELS
-from tests.utils.model_utils import (_create_detector_inputs,
-                                     _get_detector_cfg, _setup_seed)
+from mmdet3d.testing import (create_detector_inputs, get_detector_cfg,
+                             setup_seed)
 
 
 class TestPartA2(unittest.TestCase):
@@ -15,18 +15,17 @@ class TestPartA2(unittest.TestCase):
 
         assert hasattr(mmdet3d.models, 'PartA2')
         DefaultScope.get_instance('test_parta2', scope_name='mmdet3d')
-        _setup_seed(0)
-        parta2_cfg = _get_detector_cfg(
-            'parta2/PartA2_hv_secfpn_8xb2-cyclic-80e_kitti-3d-3class.py')
+        setup_seed(0)
+        parta2_cfg = get_detector_cfg(
+            'parta2/parta2_hv_secfpn_8xb2-cyclic-80e_kitti-3d-3class.py')
         model = MODELS.build(parta2_cfg)
         num_gt_instance = 2
-        packed_inputs = _create_detector_inputs(
-            num_gt_instance=num_gt_instance)
+        packed_inputs = create_detector_inputs(num_gt_instance=num_gt_instance)
 
         # TODO: Support aug data test
         # aug_packed_inputs = [
-        #     _create_detector_inputs(num_gt_instance=num_gt_instance),
-        #     _create_detector_inputs(num_gt_instance=num_gt_instance + 1)
+        #     create_detector_inputs(num_gt_instance=num_gt_instance),
+        #     create_detector_inputs(num_gt_instance=num_gt_instance + 1)
         # ]
         # test_aug_test
         # metainfo = {

tests/test_models/test_detectors/test_pointrcnn.py

@@ -4,8 +4,8 @@ import torch
 from mmengine import DefaultScope
 
 from mmdet3d.registry import MODELS
-from tests.utils.model_utils import (_create_detector_inputs,
-                                     _get_detector_cfg, _setup_seed)
+from mmdet3d.testing import (create_detector_inputs, get_detector_cfg,
+                             setup_seed)
 
 
 class TestPointRCNN(unittest.TestCase):
@@ -15,12 +15,12 @@ class TestPointRCNN(unittest.TestCase):
 
         assert hasattr(mmdet3d.models, 'PointRCNN')
         DefaultScope.get_instance('test_pointrcnn', scope_name='mmdet3d')
-        _setup_seed(0)
-        pointrcnn_cfg = _get_detector_cfg(
+        setup_seed(0)
+        pointrcnn_cfg = get_detector_cfg(
             'point_rcnn/point-rcnn_8xb2_kitti-3d-3class.py')
         model = MODELS.build(pointrcnn_cfg)
         num_gt_instance = 2
-        packed_inputs = _create_detector_inputs(
+        packed_inputs = create_detector_inputs(
             num_points=10101, num_gt_instance=num_gt_instance)
 
         if torch.cuda.is_available():

tests/test_models/test_detectors/test_pvrcnn.py

+import unittest
+
+import torch
+from mmengine import DefaultScope
+
+from mmdet3d.registry import MODELS
+from mmdet3d.testing import (create_detector_inputs, get_detector_cfg,
+                             setup_seed)
+
+
+class TestPVRCNN(unittest.TestCase):
+
+    def test_pvrcnn(self):
+        import mmdet3d.models
+
+        assert hasattr(mmdet3d.models, 'PointVoxelRCNN')
+        DefaultScope.get_instance('test_pvrcnn', scope_name='mmdet3d')
+        setup_seed(0)
+        pvrcnn_cfg = get_detector_cfg(
+            'pv_rcnn/pv_rcnn_8xb2-80e_kitti-3d-3class.py')
+        model = MODELS.build(pvrcnn_cfg)
+        num_gt_instance = 2
+        packed_inputs = create_detector_inputs(num_gt_instance=num_gt_instance)
+
+        # TODO: Support aug data test
+        # aug_packed_inputs = [
+        #     create_detector_inputs(num_gt_instance=num_gt_instance),
+        #     create_detector_inputs(num_gt_instance=num_gt_instance + 1)
+        # ]
+        # test_aug_test
+        # metainfo = {
+        #     'pcd_scale_factor': 1,
+        #     'pcd_horizontal_flip': 1,
+        #     'pcd_vertical_flip': 1,
+        #     'box_type_3d': LiDARInstance3DBoxes
+        # }
+        # for item in aug_packed_inputs:
+        #     for batch_id in len(item['data_samples']):
+        #         item['data_samples'][batch_id].set_metainfo(metainfo)
+
+        if torch.cuda.is_available():
+            model = model.cuda()
+            # test simple_test
+            with torch.no_grad():
+                data = model.data_preprocessor(packed_inputs, True)
+                torch.cuda.empty_cache()
+                results = model.forward(**data, mode='predict')
+            self.assertEqual(len(results), 1)
+            self.assertIn('bboxes_3d', results[0].pred_instances_3d)
+            self.assertIn('scores_3d', results[0].pred_instances_3d)
+            self.assertIn('labels_3d', results[0].pred_instances_3d)
+
+            # save the memory
+            with torch.no_grad():
+                losses = model.forward(**data, mode='loss')
+                torch.cuda.empty_cache()
+            self.assertGreater(losses['loss_rpn_cls'][0], 0)
+            self.assertGreaterEqual(losses['loss_rpn_bbox'][0], 0)
+            self.assertGreaterEqual(losses['loss_rpn_dir'][0], 0)
+            self.assertGreater(losses['loss_semantic'], 0)
+            self.assertGreaterEqual(losses['loss_bbox'], 0)
+            self.assertGreaterEqual(losses['loss_cls'], 0)
+            self.assertGreaterEqual(losses['loss_corner'], 0)

tests/test_models/test_detectors/test_sassd.py

+import unittest
+
+import torch
+from mmengine import DefaultScope
+
+from mmdet3d.registry import MODELS
+from mmdet3d.testing import (create_detector_inputs, get_detector_cfg,
+                             setup_seed)
+
+
+class TestSDSSD(unittest.TestCase):
+
+    def test_3dssd(self):
+        import mmdet3d.models
+
+        assert hasattr(mmdet3d.models, 'SASSD')
+        DefaultScope.get_instance('test_sassd', scope_name='mmdet3d')
+        setup_seed(0)
+        voxel_net_cfg = get_detector_cfg(
+            'sassd/sassd_8xb6-80e_kitti-3d-3class.py')
+        model = MODELS.build(voxel_net_cfg)
+        num_gt_instance = 3
+        packed_inputs = create_detector_inputs(
+            num_gt_instance=num_gt_instance, num_classes=1)
+
+        if torch.cuda.is_available():
+            model = model.cuda()
+            # test simple_test
+            with torch.no_grad():
+                data = model.data_preprocessor(packed_inputs, True)
+                torch.cuda.empty_cache()
+                results = model.forward(**data, mode='predict')
+            self.assertEqual(len(results), 1)
+            self.assertIn('bboxes_3d', results[0].pred_instances_3d)
+            self.assertIn('scores_3d', results[0].pred_instances_3d)
+            self.assertIn('labels_3d', results[0].pred_instances_3d)
+
+            losses = model.forward(**data, mode='loss')
+            self.assertGreaterEqual(losses['loss_dir'][0], 0)
+            self.assertGreaterEqual(losses['loss_bbox'][0], 0)
+            self.assertGreaterEqual(losses['loss_cls'][0], 0)
+            self.assertGreater(losses['aux_loss_cls'][0], 0)
+            self.assertGreater(losses['aux_loss_reg'][0], 0)

tests/test_models/test_detectors/test_votenet.py

@@ -4,8 +4,8 @@ import torch
 from mmengine import DefaultScope
 
 from mmdet3d.registry import MODELS
-from tests.utils.model_utils import (_create_detector_inputs,
-                                     _get_detector_cfg, _setup_seed)
+from mmdet3d.testing import (create_detector_inputs, get_detector_cfg,
+                             setup_seed)
 
 
 class TestVotenet(unittest.TestCase):
@@ -15,18 +15,16 @@ class TestVotenet(unittest.TestCase):
 
         assert hasattr(mmdet3d.models, 'VoteNet')
         DefaultScope.get_instance('test_vote_net', scope_name='mmdet3d')
-        _setup_seed(0)
-        voxel_net_cfg = _get_detector_cfg(
-            'votenet/votenet_8xb16_sunrgbd-3d.py')
+        setup_seed(0)
+        voxel_net_cfg = get_detector_cfg('votenet/votenet_8xb16_sunrgbd-3d.py')
         model = MODELS.build(voxel_net_cfg)
         num_gt_instance = 50
-        packed_inputs = _create_detector_inputs(
-            num_gt_instance=num_gt_instance)
+        packed_inputs = create_detector_inputs(num_gt_instance=num_gt_instance)
 
         # TODO: Support aug test
         # aug_data = [
-        #     _create_detector_inputs(num_gt_instance=num_gt_instance),
-        #     _create_detector_inputs(num_gt_instance=num_gt_instance + 1)
+        #     create_detector_inputs(num_gt_instance=num_gt_instance),
+        #     create_detector_inputs(num_gt_instance=num_gt_instance + 1)
         # ]
         # # test_aug_test
         # metainfo = {

tests/test_models/test_detectors/test_voxelnet.py

@@ -5,8 +5,8 @@ import torch
 from mmengine import DefaultScope
 
 from mmdet3d.registry import MODELS
-from tests.utils.model_utils import (_create_detector_inputs,
-                                     _get_detector_cfg, _setup_seed)
+from mmdet3d.testing import (create_detector_inputs, get_detector_cfg,
+                             setup_seed)
 
 
 class TestVoxelNet(unittest.TestCase):
@@ -16,18 +16,17 @@ class TestVoxelNet(unittest.TestCase):
 
         assert hasattr(mmdet3d.models, 'VoxelNet')
         DefaultScope.get_instance('test_voxelnet', scope_name='mmdet3d')
-        _setup_seed(0)
-        pointpillars_cfg = _get_detector_cfg(
+        setup_seed(0)
+        pointpillars_cfg = get_detector_cfg(
             'pointpillars/pointpillars_hv_secfpn_8xb6-160e_kitti-3d-3class.py')
         model = MODELS.build(pointpillars_cfg)
         num_gt_instance = 2
-        packed_inputs = _create_detector_inputs(
-            num_gt_instance=num_gt_instance)
+        packed_inputs = create_detector_inputs(num_gt_instance=num_gt_instance)
 
         # TODO: Support aug_test
         # aug_data = [
-        #     _create_detector_inputs(num_gt_instance=num_gt_instance),
-        #     _create_detector_inputs(num_gt_instance=num_gt_instance + 1)
+        #     create_detector_inputs(num_gt_instance=num_gt_instance),
+        #     create_detector_inputs(num_gt_instance=num_gt_instance + 1)
         # ]
         # # test_aug_test
         # metainfo = {

tests/test_models/test_losses/test_rotated_iou_loss.py

+# Copyright (c) OpenMMLab. All rights reserved.
+
+import numpy as np
+import torch
+
+from mmdet3d.models.losses import RotatedIoU3DLoss
+
+
+def test_rotated_iou_3d_loss():
+
+    if not torch.cuda.is_available():
+        return
+
+    boxes1 = torch.tensor([[.5, .5, .5, 1., 1., 1., .0],
+                           [.5, .5, .5, 1., 1., 1., .0],
+                           [.5, .5, .5, 1., 1., 1., .0],
+                           [.5, .5, .5, 1., 1., 1., .0],
+                           [.5, .5, .5, 1., 1., 1., .0]]).cuda()
+    boxes2 = torch.tensor([[.5, .5, .5, 1., 1., 1., .0],
+                           [.5, .5, .5, 1., 1., 2., np.pi / 2],
+                           [.5, .5, .5, 1., 1., 1., np.pi / 4],
+                           [1., 1., 1., 1., 1., 1., .0],
+                           [-1.5, -1.5, -1.5, 2.5, 2.5, 2.5, .0]]).cuda()
+
+    expect_ious = 1 - torch.tensor([[1., .5, .7071, 1 / 15, .0]]).cuda()
+    ious = RotatedIoU3DLoss(reduction='none')(boxes1, boxes2)
+    assert torch.allclose(ious, expect_ious, atol=1e-4)

tests/test_models/test_necks/test_second_fpn.py

@@ -45,8 +45,8 @@ def test_centerpoint_fpn():
 
     second_cfg = dict(
         type='SECOND',
-        in_channels=64,
-        out_channels=[64, 128, 256],
+        in_channels=2,
+        out_channels=[2, 2, 2],
         layer_nums=[3, 5, 5],
         layer_strides=[2, 2, 2],
         norm_cfg=dict(type='BN', eps=1e-3, momentum=0.01),
@@ -57,8 +57,8 @@ def test_centerpoint_fpn():
     # centerpoint usage of fpn
     centerpoint_fpn_cfg = dict(
         type='SECONDFPN',
-        in_channels=[64, 128, 256],
-        out_channels=[128, 128, 128],
+        in_channels=[2, 2, 2],
+        out_channels=[2, 2, 2],
         upsample_strides=[0.5, 1, 2],
         norm_cfg=dict(type='BN', eps=1e-3, momentum=0.01),
         upsample_cfg=dict(type='deconv', bias=False),
@@ -67,17 +67,17 @@ def test_centerpoint_fpn():
     # original usage of fpn
     fpn_cfg = dict(
         type='SECONDFPN',
-        in_channels=[64, 128, 256],
+        in_channels=[2, 2, 2],
         upsample_strides=[1, 2, 4],
-        out_channels=[128, 128, 128])
+        out_channels=[2, 2, 2])
 
     second_fpn = build_neck(fpn_cfg)
 
     centerpoint_second_fpn = build_neck(centerpoint_fpn_cfg)
 
-    input = torch.rand([4, 64, 512, 512])
+    input = torch.rand([2, 2, 32, 32])
     sec_output = second(input)
     centerpoint_output = centerpoint_second_fpn(sec_output)
     second_output = second_fpn(sec_output)
-    assert centerpoint_output[0].shape == torch.Size([4, 384, 128, 128])
-    assert second_output[0].shape == torch.Size([4, 384, 256, 256])
+    assert centerpoint_output[0].shape == torch.Size([2, 6, 8, 8])
+    assert second_output[0].shape == torch.Size([2, 6, 16, 16])

tests/test_models/test_task_modules/test_anchor/test_anchor_3d_generator.py

@@ -45,10 +45,10 @@ def test_anchor_3d_range_generator():
                         '\nrotations=[0, 1.57],\nreshape_out=False,' \
                         '\nsize_per_range=True)'
     assert repr_str == expected_repr_str
-    featmap_size = (256, 256)
+    featmap_size = (8, 8)
     mr_anchors = anchor_generator.single_level_grid_anchors(
         featmap_size, 1.1, device=device)
-    assert mr_anchors.shape == torch.Size([1, 256, 256, 3, 2, 7])
+    assert mr_anchors.shape == torch.Size([1, 8, 8, 3, 2, 7])
 
 
 def test_aligned_anchor_generator():
@@ -80,108 +80,108 @@ def test_aligned_anchor_generator():
         size_per_range=False,
         reshape_out=True)
 
-    featmap_sizes = [(256, 256), (128, 128), (64, 64)]
+    featmap_sizes = [(16, 16), (8, 8), (4, 4)]
     anchor_generator = TASK_UTILS.build(anchor_generator_cfg)
     assert anchor_generator.num_base_anchors == 8
 
     # check base anchors
     expected_grid_anchors = [
         torch.tensor([[
-            -51.0000, -51.0000, -1.8000, 2.5981, 0.8660, 1.0000, 0.0000,
+            -48.0000, -48.0000, -1.8000, 2.5981, 0.8660, 1.0000, 0.0000,
             0.0000, 0.0000
         ],
                       [
-                          -51.0000, -51.0000, -1.8000, 0.4000, 0.4000, 1.0000,
+                          -48.0000, -48.0000, -1.8000, 0.4000, 0.4000, 1.0000,
                           1.5700, 0.0000, 0.0000
                       ],
                       [
-                          -50.6000, -51.0000, -1.8000, 0.4000, 0.4000, 1.0000,
+                          -41.6000, -48.0000, -1.8000, 0.4000, 0.4000, 1.0000,
                           0.0000, 0.0000, 0.0000
                       ],
                       [
-                          -50.2000, -51.0000, -1.8000, 1.0000, 1.0000, 1.0000,
+                          -35.2000, -48.0000, -1.8000, 1.0000, 1.0000, 1.0000,
                           1.5700, 0.0000, 0.0000
                       ],
                       [
-                          -49.8000, -51.0000, -1.8000, 1.0000, 1.0000, 1.0000,
+                          -28.8000, -48.0000, -1.8000, 1.0000, 1.0000, 1.0000,
                           0.0000, 0.0000, 0.0000
                       ],
                       [
-                          -49.4000, -51.0000, -1.8000, 1.7321, 0.5774, 1.0000,
+                          -22.4000, -48.0000, -1.8000, 1.7321, 0.5774, 1.0000,
                           1.5700, 0.0000, 0.0000
                       ],
                       [
-                          -49.0000, -51.0000, -1.8000, 1.7321, 0.5774, 1.0000,
+                          -16.0000, -48.0000, -1.8000, 1.7321, 0.5774, 1.0000,
                           0.0000, 0.0000, 0.0000
                       ],
                       [
-                          -48.6000, -51.0000, -1.8000, 2.5981, 0.8660, 1.0000,
+                          -9.6000, -48.0000, -1.8000, 2.5981, 0.8660, 1.0000,
                           1.5700, 0.0000, 0.0000
                       ]],
                      device=device),
         torch.tensor([[
-            -50.8000, -50.8000, -1.8000, 5.1962, 1.7320, 2.0000, 0.0000,
+            -44.8000, -44.8000, -1.8000, 5.1962, 1.7320, 2.0000, 0.0000,
             0.0000, 0.0000
         ],
                       [
-                          -50.8000, -50.8000, -1.8000, 0.8000, 0.8000, 2.0000,
+                          -44.8000, -44.8000, -1.8000, 0.8000, 0.8000, 2.0000,
                           1.5700, 0.0000, 0.0000
                       ],
                       [
-                          -50.0000, -50.8000, -1.8000, 0.8000, 0.8000, 2.0000,
+                          -32.0000, -44.8000, -1.8000, 0.8000, 0.8000, 2.0000,
                           0.0000, 0.0000, 0.0000
                       ],
                       [
-                          -49.2000, -50.8000, -1.8000, 2.0000, 2.0000, 2.0000,
+                          -19.2000, -44.8000, -1.8000, 2.0000, 2.0000, 2.0000,
                           1.5700, 0.0000, 0.0000
                       ],
                       [
-                          -48.4000, -50.8000, -1.8000, 2.0000, 2.0000, 2.0000,
+                          -6.4000, -44.8000, -1.8000, 2.0000, 2.0000, 2.0000,
                           0.0000, 0.0000, 0.0000
                       ],
                       [
-                          -47.6000, -50.8000, -1.8000, 3.4642, 1.1548, 2.0000,
+                          6.4000, -44.8000, -1.8000, 3.4642, 1.1548, 2.0000,
                           1.5700, 0.0000, 0.0000
                       ],
                       [
-                          -46.8000, -50.8000, -1.8000, 3.4642, 1.1548, 2.0000,
+                          19.2000, -44.8000, -1.8000, 3.4642, 1.1548, 2.0000,
                           0.0000, 0.0000, 0.0000
                       ],
                       [
-                          -46.0000, -50.8000, -1.8000, 5.1962, 1.7320, 2.0000,
+                          32.0000, -44.8000, -1.8000, 5.1962, 1.7320, 2.0000,
                           1.5700, 0.0000, 0.0000
                       ]],
                      device=device),
         torch.tensor([[
-            -50.4000, -50.4000, -1.8000, 10.3924, 3.4640, 4.0000, 0.0000,
+            -38.4000, -38.4000, -1.8000, 10.3924, 3.4640, 4.0000, 0.0000,
             0.0000, 0.0000
         ],
                       [
-                          -50.4000, -50.4000, -1.8000, 1.6000, 1.6000, 4.0000,
+                          -38.4000, -38.4000, -1.8000, 1.6000, 1.6000, 4.0000,
                           1.5700, 0.0000, 0.0000
                       ],
                       [
-                          -48.8000, -50.4000, -1.8000, 1.6000, 1.6000, 4.0000,
+                          -12.8000, -38.4000, -1.8000, 1.6000, 1.6000, 4.0000,
                           0.0000, 0.0000, 0.0000
                       ],
                       [
-                          -47.2000, -50.4000, -1.8000, 4.0000, 4.0000, 4.0000,
+                          12.8000, -38.4000, -1.8000, 4.0000, 4.0000, 4.0000,
                           1.5700, 0.0000, 0.0000
                       ],
                       [
-                          -45.6000, -50.4000, -1.8000, 4.0000, 4.0000, 4.0000,
+                          38.4000, -38.4000, -1.8000, 4.0000, 4.0000, 4.0000,
                           0.0000, 0.0000, 0.0000
                       ],
                       [
-                          -44.0000, -50.4000, -1.8000, 6.9284, 2.3096, 4.0000,
+                          -38.4000, -12.8000, -1.8000, 6.9284, 2.3096, 4.0000,
                           1.5700, 0.0000, 0.0000
                       ],
                       [
-                          -42.4000, -50.4000, -1.8000, 6.9284, 2.3096, 4.0000,
+                          -12.8000, -12.8000, -1.8000, 6.9284, 2.3096, 4.0000,
                           0.0000, 0.0000, 0.0000
                       ],
                       [
-                          -40.8000, -50.4000, -1.8000, 10.3924, 3.4640, 4.0000,
+                          12.8000, -12.8000, -1.8000, 10.3924, 3.4640, 4.0000,
                           1.5700, 0.0000, 0.0000
                       ]],
                      device=device)
@@ -189,9 +189,9 @@ def test_aligned_anchor_generator():
     multi_level_anchors = anchor_generator.grid_anchors(
         featmap_sizes, device=device)
     expected_multi_level_shapes = [
-        torch.Size([524288, 9]),
-        torch.Size([131072, 9]),
-        torch.Size([32768, 9])
+        torch.Size([2048, 9]),
+        torch.Size([512, 9]),
+        torch.Size([128, 9])
     ]
     for i, single_level_anchor in enumerate(multi_level_anchors):
         assert single_level_anchor.shape == expected_multi_level_shapes[i]

tests/test_models/test_task_modules/test_voxel/test_voxel_generator.py

@@ -6,17 +6,15 @@ from mmdet3d.models.task_modules.voxel import VoxelGenerator
 
 def test_voxel_generator():
     np.random.seed(0)
-    voxel_size = [0.5, 0.5, 0.5]
-    point_cloud_range = [0, -40, -3, 70.4, 40, 1]
-    max_num_points = 1000
+    voxel_size = [5, 5, 1]
+    point_cloud_range = [0, 0, 0, 20, 40, 4]
+    max_num_points = 5
     self = VoxelGenerator(voxel_size, point_cloud_range, max_num_points)
-    points = np.random.rand(1000, 4)
+    points = np.random.uniform(0, 4, (20, 3))
     voxels = self.generate(points)
     voxels, coors, num_points_per_voxel = voxels
-    expected_coors = np.array([[7, 81, 1], [6, 81, 0], [7, 80, 1], [6, 81, 1],
-                               [7, 81, 0], [6, 80, 1], [7, 80, 0], [6, 80, 0]])
-    expected_num_points_per_voxel = np.array(
-        [120, 121, 127, 134, 115, 127, 125, 131])
-    assert voxels.shape == (8, 1000, 4)
+    expected_coors = np.array([[2, 0, 0], [3, 0, 0], [0, 0, 0], [1, 0, 0]])
+    expected_num_points_per_voxel = np.array([5, 5, 5, 3])
+    assert voxels.shape == (4, 5, 3)
     assert np.all(coors == expected_coors)
     assert np.all(num_points_per_voxel == expected_num_points_per_voxel)

tools/create_data.py

@@ -35,9 +35,6 @@ def kitti_data_prep(root_path,
     info_train_path = osp.join(out_dir, f'{info_prefix}_infos_train.pkl')
     info_val_path = osp.join(out_dir, f'{info_prefix}_infos_val.pkl')
     info_trainval_path = osp.join(out_dir, f'{info_prefix}_infos_trainval.pkl')
-    kitti.export_2d_annotation(root_path, info_train_path)
-    kitti.export_2d_annotation(root_path, info_val_path)
-    kitti.export_2d_annotation(root_path, info_trainval_path)
     update_pkl_infos('kitti', out_dir=out_dir, pkl_path=info_train_path)
     update_pkl_infos('kitti', out_dir=out_dir, pkl_path=info_val_path)
     update_pkl_infos('kitti', out_dir=out_dir, pkl_path=info_trainval_path)
@@ -76,17 +73,11 @@ def nuscenes_data_prep(root_path,
 
     if version == 'v1.0-test':
         info_test_path = osp.join(out_dir, f'{info_prefix}_infos_test.pkl')
-        nuscenes_converter.export_2d_annotation(
-            root_path, info_test_path, version=version)
         update_pkl_infos('nuscenes', out_dir=out_dir, pkl_path=info_test_path)
         return
 
     info_train_path = osp.join(out_dir, f'{info_prefix}_infos_train.pkl')
     info_val_path = osp.join(out_dir, f'{info_prefix}_infos_val.pkl')
-    nuscenes_converter.export_2d_annotation(
-        root_path, info_train_path, version=version)
-    nuscenes_converter.export_2d_annotation(
-        root_path, info_val_path, version=version)
     update_pkl_infos('nuscenes', out_dir=out_dir, pkl_path=info_train_path)
     update_pkl_infos('nuscenes', out_dir=out_dir, pkl_path=info_val_path)
     create_groundtruth_database(dataset_name, root_path, info_prefix,
@@ -265,6 +256,11 @@ args = parser.parse_args()
 if __name__ == '__main__':
     from mmdet3d.utils import register_all_modules
     register_all_modules()
+
+    # Set to spawn mode to avoid stuck when process dataset creating
+    import multiprocessing
+    multiprocessing.set_start_method('spawn')
+
     if args.dataset == 'kitti':
         kitti_data_prep(
             root_path=args.root_path,

tools/dataset_converters/create_gt_database.py

@@ -6,13 +6,13 @@ import mmcv
 import mmengine
 import numpy as np
 from mmcv.ops import roi_align
+from mmdet.evaluation import bbox_overlaps
 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.structures.ops import box_np_ops as box_np_ops
-from mmdet.evaluation import bbox_overlaps
 
 
 def _poly2mask(mask_ann, img_h, img_w):
@@ -243,7 +243,7 @@ def create_groundtruth_database(dataset_class_name,
         image_idx = example['sample_idx']
         points = example['points'].tensor.numpy()
         gt_boxes_3d = annos['gt_bboxes_3d'].tensor.numpy()
-        names = [dataset.metainfo['CLASSES'][i] for i in annos['gt_labels_3d']]
+        names = [dataset.metainfo['classes'][i] for i in annos['gt_labels_3d']]
         group_dict = dict()
         if 'group_ids' in annos:
             group_ids = annos['group_ids']
@@ -409,7 +409,7 @@ class GTDatabaseCreater:
         points = example['points'].tensor.numpy()
         gt_boxes_3d = annos['gt_bboxes_3d'].tensor.numpy()
         names = [
-            self.dataset.metainfo['CLASSES'][i] for i in annos['gt_labels_3d']
+            self.dataset.metainfo['classes'][i] for i in annos['gt_labels_3d']
         ]
         group_dict = dict()
         if 'group_ids' in annos:

tools/dataset_converters/lyft_converter.py

@@ -131,6 +131,7 @@ def _fill_trainval_infos(lyft,
 
         info = {
             'lidar_path': lidar_path,
+            'num_features': 5,
             'token': sample['token'],
             'sweeps': [],
             'cams': dict(),

tools/dataset_converters/s3dis_data_utils.py

@@ -129,7 +129,7 @@ class S3DISData(object):
                 - gt_num (int): Number of boxes.
         """
         bboxes, labels = [], []
-        for i in range(1, pts_instance_mask.max()):
+        for i in range(1, pts_instance_mask.max() + 1):
             ids = pts_instance_mask == i
             # check if all instance points have same semantic label
             assert pts_semantic_mask[ids].min() == pts_semantic_mask[ids].max()

tools/dataset_converters/update_infos_to_v2.py

@@ -4,7 +4,7 @@
 Example:
     python tools/dataset_converters/update_infos_to_v2.py
         --dataset kitti
-        --pkl ./data/kitti/kitti_infos_train.pkl
+        --pkl-path ./data/kitti/kitti_infos_train.pkl
         --out-dir ./kitti_v2/
 """
 
@@ -12,6 +12,7 @@ import argparse
 import copy
 import time
 from os import path as osp
+from pathlib import Path
 
 import mmengine
 import numpy as np
@@ -80,9 +81,6 @@ def get_empty_lidar_points():
         num_pts_feats=None,
         # (str, optional): Path of LiDAR data file.
         lidar_path=None,
-        # (list[list[float]]): Transformation matrix from lidar
-        # or depth to image with shape [4, 4].
-        lidar2img=None,
         # (list[list[float]], optional): Transformation matrix
         # from lidar to ego-vehicle
         # with shape [4, 4].
@@ -120,6 +118,9 @@ def get_empty_img_info():
         # matrix from camera to image with
         # shape [3, 3], [3, 4] or [4, 4].
         cam2img=None,
+        # (list[list[float]]): Transformation matrix from lidar
+        # or depth to image with shape [4, 4].
+        lidar2img=None,
         # (list[list[float]], optional) : Transformation
         # matrix from camera to ego-vehicle
         # with shape [4, 4].
@@ -159,7 +160,7 @@ def get_empty_standard_data_info(
 
     data_info = dict(
         # (str): Sample id of the frame.
-        sample_id=None,
+        sample_idx=None,
         # (str, optional): '000010'
         token=None,
         **get_single_image_sweep(camera_types),
@@ -261,13 +262,9 @@ def update_nuscenes_infos(pkl_path, out_dir):
     print(f'Reading from input file: {pkl_path}.')
     data_list = mmengine.load(pkl_path)
     METAINFO = {
-        'CLASSES':
+        'classes':
         ('car', 'truck', 'trailer', 'bus', 'construction_vehicle', 'bicycle',
          'motorcycle', 'pedestrian', 'traffic_cone', 'barrier'),
-        'DATASET':
-        'Nuscenes',
-        'version':
-        data_list['metadata']['version']
     }
     nusc = NuScenes(
         version=data_list['metadata']['version'],
@@ -287,8 +284,8 @@ def update_nuscenes_infos(pkl_path, out_dir):
             ori_info_dict['ego2global_translation'])
         temp_data_info['lidar_points']['num_pts_feats'] = ori_info_dict.get(
             'num_features', 5)
-        temp_data_info['lidar_points']['lidar_path'] = ori_info_dict[
-            'lidar_path'].split('/')[-1]
+        temp_data_info['lidar_points']['lidar_path'] = Path(
+            ori_info_dict['lidar_path']).name
         temp_data_info['lidar_points'][
             'lidar2ego'] = convert_quaternion_to_matrix(
                 ori_info_dict['lidar2ego_rotation'],
@@ -318,8 +315,8 @@ def update_nuscenes_infos(pkl_path, out_dir):
         temp_data_info['images'] = {}
         for cam in ori_info_dict['cams']:
             empty_img_info = get_empty_img_info()
-            empty_img_info['img_path'] = ori_info_dict['cams'][cam][
-                'data_path'].split('/')[-1]
+            empty_img_info['img_path'] = Path(
+                ori_info_dict['cams'][cam]['data_path']).name
             empty_img_info['cam2img'] = ori_info_dict['cams'][cam][
                 'cam_intrinsic'].tolist()
             empty_img_info['sample_data_token'] = ori_info_dict['cams'][cam][
@@ -344,8 +341,8 @@ def update_nuscenes_infos(pkl_path, out_dir):
             empty_instance = get_empty_instance()
             empty_instance['bbox_3d'] = ori_info_dict['gt_boxes'][
                 i, :].tolist()
-            if ori_info_dict['gt_names'][i] in METAINFO['CLASSES']:
-                empty_instance['bbox_label'] = METAINFO['CLASSES'].index(
+            if ori_info_dict['gt_names'][i] in METAINFO['classes']:
+                empty_instance['bbox_label'] = METAINFO['classes'].index(
                     ori_info_dict['gt_names'][i])
             else:
                 ignore_class_name.add(ori_info_dict['gt_names'][i])
@@ -363,11 +360,20 @@ def update_nuscenes_infos(pkl_path, out_dir):
             ori_info_dict, nusc)
         temp_data_info, _ = clear_data_info_unused_keys(temp_data_info)
         converted_list.append(temp_data_info)
-    pkl_name = pkl_path.split('/')[-1]
+    pkl_name = Path(pkl_path).name
     out_path = osp.join(out_dir, pkl_name)
     print(f'Writing to output file: {out_path}.')
     print(f'ignore classes: {ignore_class_name}')
-    converted_data_info = dict(metainfo=METAINFO, data_list=converted_list)
+
+    metainfo = dict()
+    metainfo['categories'] = {k: i for i, k in enumerate(METAINFO['classes'])}
+    if ignore_class_name:
+        for ignore_class in ignore_class_name:
+            metainfo['categories'][ignore_class] = -1
+    metainfo['dataset'] = 'nuscenes'
+    metainfo['version'] = data_list['metadata']['version']
+    metainfo['info_version'] = '1.1'
+    converted_data_info = dict(metainfo=metainfo, data_list=converted_list)
 
     mmengine.dump(converted_data_info, out_path, 'pkl')
 
@@ -381,7 +387,7 @@ def update_kitti_infos(pkl_path, out_dir):
     # TODO update to full label
     # TODO discuss how to process 'Van', 'DontCare'
     METAINFO = {
-        'CLASSES': ('Pedestrian', 'Cyclist', 'Car', 'Van', 'Truck',
+        'classes': ('Pedestrian', 'Cyclist', 'Car', 'Van', 'Truck',
                     'Person_sitting', 'Tram', 'Misc'),
     }
     print(f'Reading from input file: {pkl_path}.')
@@ -405,15 +411,15 @@ def update_kitti_infos(pkl_path, out_dir):
         temp_data_info['images']['CAM3']['cam2img'] = ori_info_dict['calib'][
             'P3'].tolist()
 
-        temp_data_info['images']['CAM2']['img_path'] = ori_info_dict['image'][
-            'image_path'].split('/')[-1]
+        temp_data_info['images']['CAM2']['img_path'] = Path(
+            ori_info_dict['image']['image_path']).name
         h, w = ori_info_dict['image']['image_shape']
         temp_data_info['images']['CAM2']['height'] = h
         temp_data_info['images']['CAM2']['width'] = w
         temp_data_info['lidar_points']['num_pts_feats'] = ori_info_dict[
             'point_cloud']['num_features']
-        temp_data_info['lidar_points']['lidar_path'] = ori_info_dict[
-            'point_cloud']['velodyne_path'].split('/')[-1]
+        temp_data_info['lidar_points']['lidar_path'] = Path(
+            ori_info_dict['point_cloud']['velodyne_path']).name
 
         rect = ori_info_dict['calib']['R0_rect'].astype(np.float32)
         Trv2c = ori_info_dict['calib']['Tr_velo_to_cam'].astype(np.float32)
@@ -446,8 +452,8 @@ def update_kitti_infos(pkl_path, out_dir):
             empty_instance = get_empty_instance()
             empty_instance['bbox'] = anns['bbox'][instance_id].tolist()
 
-            if anns['name'][instance_id] in METAINFO['CLASSES']:
-                empty_instance['bbox_label'] = METAINFO['CLASSES'].index(
+            if anns['name'][instance_id] in METAINFO['classes']:
+                empty_instance['bbox_label'] = METAINFO['classes'].index(
                     anns['name'][instance_id])
             else:
                 ignore_class_name.add(anns['name'][instance_id])
@@ -493,12 +499,20 @@ def update_kitti_infos(pkl_path, out_dir):
         temp_data_info['cam_instances'] = cam_instances
         temp_data_info, _ = clear_data_info_unused_keys(temp_data_info)
         converted_list.append(temp_data_info)
-    pkl_name = pkl_path.split('/')[-1]
+    pkl_name = Path(pkl_path).name
     out_path = osp.join(out_dir, pkl_name)
     print(f'Writing to output file: {out_path}.')
     print(f'ignore classes: {ignore_class_name}')
-    converted_data_info = dict(
-        metainfo={'DATASET': 'KITTI'}, data_list=converted_list)
+
+    # dataset metainfo
+    metainfo = dict()
+    metainfo['categories'] = {k: i for i, k in enumerate(METAINFO['classes'])}
+    if ignore_class_name:
+        for ignore_class in ignore_class_name:
+            metainfo['categories'][ignore_class] = -1
+    metainfo['dataset'] = 'kitti'
+    metainfo['info_version'] = '1.1'
+    converted_data_info = dict(metainfo=metainfo, data_list=converted_list)
 
     mmengine.dump(converted_data_info, out_path, 'pkl')
 
@@ -509,7 +523,7 @@ def update_s3dis_infos(pkl_path, out_dir):
         print(f'Warning, you may overwriting '
               f'the original data {pkl_path}.')
         time.sleep(5)
-    METAINFO = {'CLASSES': ('table', 'chair', 'sofa', 'bookcase', 'board')}
+    METAINFO = {'classes': ('table', 'chair', 'sofa', 'bookcase', 'board')}
     print(f'Reading from input file: {pkl_path}.')
     data_list = mmengine.load(pkl_path)
     print('Start updating:')
@@ -519,12 +533,12 @@ def update_s3dis_infos(pkl_path, out_dir):
         temp_data_info['sample_idx'] = i
         temp_data_info['lidar_points']['num_pts_feats'] = ori_info_dict[
             'point_cloud']['num_features']
-        temp_data_info['lidar_points']['lidar_path'] = ori_info_dict[
-            'pts_path'].split('/')[-1]
-        temp_data_info['pts_semantic_mask_path'] = ori_info_dict[
-            'pts_semantic_mask_path'].split('/')[-1]
-        temp_data_info['pts_instance_mask_path'] = ori_info_dict[
-            'pts_instance_mask_path'].split('/')[-1]
+        temp_data_info['lidar_points']['lidar_path'] = Path(
+            ori_info_dict['pts_path']).name
+        temp_data_info['pts_semantic_mask_path'] = Path(
+            ori_info_dict['pts_semantic_mask_path']).name
+        temp_data_info['pts_instance_mask_path'] = Path(
+            ori_info_dict['pts_instance_mask_path']).name
 
         # TODO support camera
         # np.linalg.inv(info['axis_align_matrix'] @ extrinsic): depth2cam
@@ -541,12 +555,12 @@ def update_s3dis_infos(pkl_path, out_dir):
                     empty_instance['bbox_3d'] = anns['gt_boxes_upright_depth'][
                         instance_id].tolist()
 
-                    if anns['class'][instance_id] < len(METAINFO['CLASSES']):
+                    if anns['class'][instance_id] < len(METAINFO['classes']):
                         empty_instance['bbox_label_3d'] = anns['class'][
                             instance_id]
                     else:
                         ignore_class_name.add(
-                            METAINFO['CLASSES'][anns['class'][instance_id]])
+                            METAINFO['classes'][anns['class'][instance_id]])
                         empty_instance['bbox_label_3d'] = -1
 
                     empty_instance = clear_instance_unused_keys(empty_instance)
@@ -554,12 +568,21 @@ def update_s3dis_infos(pkl_path, out_dir):
             temp_data_info['instances'] = instance_list
         temp_data_info, _ = clear_data_info_unused_keys(temp_data_info)
         converted_list.append(temp_data_info)
-    pkl_name = pkl_path.split('/')[-1]
+    pkl_name = Path(pkl_path).name
     out_path = osp.join(out_dir, pkl_name)
     print(f'Writing to output file: {out_path}.')
     print(f'ignore classes: {ignore_class_name}')
-    converted_data_info = dict(
-        metainfo={'DATASET': 'S3DIS'}, data_list=converted_list)
+
+    # dataset metainfo
+    metainfo = dict()
+    metainfo['categories'] = {k: i for i, k in enumerate(METAINFO['classes'])}
+    if ignore_class_name:
+        for ignore_class in ignore_class_name:
+            metainfo['categories'][ignore_class] = -1
+    metainfo['dataset'] = 's3dis'
+    metainfo['info_version'] = '1.1'
+
+    converted_data_info = dict(metainfo=metainfo, data_list=converted_list)
 
     mmengine.dump(converted_data_info, out_path, 'pkl')
 
@@ -571,7 +594,7 @@ def update_scannet_infos(pkl_path, out_dir):
               f'the original data {pkl_path}.')
         time.sleep(5)
     METAINFO = {
-        'CLASSES':
+        'classes':
         ('cabinet', 'bed', 'chair', 'sofa', 'table', 'door', 'window',
          'bookshelf', 'picture', 'counter', 'desk', 'curtain', 'refrigerator',
          'showercurtrain', 'toilet', 'sink', 'bathtub', 'garbagebin')
@@ -584,12 +607,12 @@ def update_scannet_infos(pkl_path, out_dir):
         temp_data_info = get_empty_standard_data_info()
         temp_data_info['lidar_points']['num_pts_feats'] = ori_info_dict[
             'point_cloud']['num_features']
-        temp_data_info['lidar_points']['lidar_path'] = ori_info_dict[
-            'pts_path'].split('/')[-1]
-        temp_data_info['pts_semantic_mask_path'] = ori_info_dict[
-            'pts_semantic_mask_path'].split('/')[-1]
-        temp_data_info['pts_instance_mask_path'] = ori_info_dict[
-            'pts_instance_mask_path'].split('/')[-1]
+        temp_data_info['lidar_points']['lidar_path'] = Path(
+            ori_info_dict['pts_path']).name
+        temp_data_info['pts_semantic_mask_path'] = Path(
+            ori_info_dict['pts_semantic_mask_path']).name
+        temp_data_info['pts_instance_mask_path'] = Path(
+            ori_info_dict['pts_instance_mask_path']).name
 
         # TODO support camera
         # np.linalg.inv(info['axis_align_matrix'] @ extrinsic): depth2cam
@@ -607,9 +630,9 @@ def update_scannet_infos(pkl_path, out_dir):
                 empty_instance['bbox_3d'] = anns['gt_boxes_upright_depth'][
                     instance_id].tolist()
 
-                if anns['name'][instance_id] in METAINFO['CLASSES']:
+                if anns['name'][instance_id] in METAINFO['classes']:
                     empty_instance['bbox_label_3d'] = METAINFO[
-                        'CLASSES'].index(anns['name'][instance_id])
+                        'classes'].index(anns['name'][instance_id])
                 else:
                     ignore_class_name.add(anns['name'][instance_id])
                     empty_instance['bbox_label_3d'] = -1
@@ -619,12 +642,21 @@ def update_scannet_infos(pkl_path, out_dir):
         temp_data_info['instances'] = instance_list
         temp_data_info, _ = clear_data_info_unused_keys(temp_data_info)
         converted_list.append(temp_data_info)
-    pkl_name = pkl_path.split('/')[-1]
+    pkl_name = Path(pkl_path).name
     out_path = osp.join(out_dir, pkl_name)
     print(f'Writing to output file: {out_path}.')
     print(f'ignore classes: {ignore_class_name}')
-    converted_data_info = dict(
-        metainfo={'DATASET': 'SCANNET'}, data_list=converted_list)
+
+    # dataset metainfo
+    metainfo = dict()
+    metainfo['categories'] = {k: i for i, k in enumerate(METAINFO['classes'])}
+    if ignore_class_name:
+        for ignore_class in ignore_class_name:
+            metainfo['categories'][ignore_class] = -1
+    metainfo['dataset'] = 'scannet'
+    metainfo['info_version'] = '1.1'
+
+    converted_data_info = dict(metainfo=metainfo, data_list=converted_list)
 
     mmengine.dump(converted_data_info, out_path, 'pkl')
 
@@ -636,7 +668,7 @@ def update_sunrgbd_infos(pkl_path, out_dir):
               f'the original data {pkl_path}.')
         time.sleep(5)
     METAINFO = {
-        'CLASSES': ('bed', 'table', 'sofa', 'chair', 'toilet', 'desk',
+        'classes': ('bed', 'table', 'sofa', 'chair', 'toilet', 'desk',
                     'dresser', 'night_stand', 'bookshelf', 'bathtub')
     }
     print(f'Reading from input file: {pkl_path}.')
@@ -647,8 +679,8 @@ def update_sunrgbd_infos(pkl_path, out_dir):
         temp_data_info = get_empty_standard_data_info()
         temp_data_info['lidar_points']['num_pts_feats'] = ori_info_dict[
             'point_cloud']['num_features']
-        temp_data_info['lidar_points']['lidar_path'] = ori_info_dict[
-            'pts_path'].split('/')[-1]
+        temp_data_info['lidar_points']['lidar_path'] = Path(
+            ori_info_dict['pts_path']).name
         calib = ori_info_dict['calib']
         rt_mat = calib['Rt']
         # follow Coord3DMode.convert_point
@@ -656,8 +688,8 @@ def update_sunrgbd_infos(pkl_path, out_dir):
                            ]) @ rt_mat.transpose(1, 0)
         depth2img = calib['K'] @ rt_mat
         temp_data_info['images']['CAM0']['depth2img'] = depth2img.tolist()
-        temp_data_info['images']['CAM0']['img_path'] = ori_info_dict['image'][
-            'image_path'].split('/')[-1]
+        temp_data_info['images']['CAM0']['img_path'] = Path(
+            ori_info_dict['image']['image_path']).name
         h, w = ori_info_dict['image']['image_shape']
         temp_data_info['images']['CAM0']['height'] = h
         temp_data_info['images']['CAM0']['width'] = w
@@ -674,9 +706,9 @@ def update_sunrgbd_infos(pkl_path, out_dir):
                 empty_instance['bbox_3d'] = anns['gt_boxes_upright_depth'][
                     instance_id].tolist()
                 empty_instance['bbox'] = anns['bbox'][instance_id].tolist()
-                if anns['name'][instance_id] in METAINFO['CLASSES']:
+                if anns['name'][instance_id] in METAINFO['classes']:
                     empty_instance['bbox_label_3d'] = METAINFO[
-                        'CLASSES'].index(anns['name'][instance_id])
+                        'classes'].index(anns['name'][instance_id])
                     empty_instance['bbox_label'] = empty_instance[
                         'bbox_label_3d']
                 else:
@@ -688,12 +720,21 @@ def update_sunrgbd_infos(pkl_path, out_dir):
         temp_data_info['instances'] = instance_list
         temp_data_info, _ = clear_data_info_unused_keys(temp_data_info)
         converted_list.append(temp_data_info)
-    pkl_name = pkl_path.split('/')[-1]
+    pkl_name = Path(pkl_path).name
     out_path = osp.join(out_dir, pkl_name)
     print(f'Writing to output file: {out_path}.')
     print(f'ignore classes: {ignore_class_name}')
-    converted_data_info = dict(
-        metainfo={'DATASET': 'SUNRGBD'}, data_list=converted_list)
+
+    # dataset metainfo
+    metainfo = dict()
+    metainfo['categories'] = {k: i for i, k in enumerate(METAINFO['classes'])}
+    if ignore_class_name:
+        for ignore_class in ignore_class_name:
+            metainfo['categories'][ignore_class] = -1
+    metainfo['dataset'] = 'sunrgbd'
+    metainfo['info_version'] = '1.1'
+
+    converted_data_info = dict(metainfo=metainfo, data_list=converted_list)
 
     mmengine.dump(converted_data_info, out_path, 'pkl')
 
@@ -706,13 +747,9 @@ def update_lyft_infos(pkl_path, out_dir):
     print(f'Reading from input file: {pkl_path}.')
     data_list = mmengine.load(pkl_path)
     METAINFO = {
-        'CLASSES':
+        'classes':
         ('car', 'truck', 'bus', 'emergency_vehicle', 'other_vehicle',
          'motorcycle', 'bicycle', 'pedestrian', 'animal'),
-        'DATASET':
-        'Nuscenes',
-        'version':
-        data_list['metadata']['version']
     }
     print('Start updating:')
     converted_list = []
@@ -724,8 +761,10 @@ def update_lyft_infos(pkl_path, out_dir):
         temp_data_info['ego2global'] = convert_quaternion_to_matrix(
             ori_info_dict['ego2global_rotation'],
             ori_info_dict['ego2global_translation'])
-        temp_data_info['lidar_points']['lidar_path'] = ori_info_dict[
-            'lidar_path'].split('/')[-1]
+        temp_data_info['lidar_points']['num_pts_feats'] = ori_info_dict.get(
+            'num_features', 5)
+        temp_data_info['lidar_points']['lidar_path'] = Path(
+            ori_info_dict['lidar_path']).name
         temp_data_info['lidar_points'][
             'lidar2ego'] = convert_quaternion_to_matrix(
                 ori_info_dict['lidar2ego_rotation'],
@@ -756,8 +795,8 @@ def update_lyft_infos(pkl_path, out_dir):
         temp_data_info['images'] = {}
         for cam in ori_info_dict['cams']:
             empty_img_info = get_empty_img_info()
-            empty_img_info['img_path'] = ori_info_dict['cams'][cam][
-                'data_path'].split('/')[-1]
+            empty_img_info['img_path'] = Path(
+                ori_info_dict['cams'][cam]['data_path']).name
             empty_img_info['cam2img'] = ori_info_dict['cams'][cam][
                 'cam_intrinsic'].tolist()
             empty_img_info['sample_data_token'] = ori_info_dict['cams'][cam][
@@ -781,8 +820,8 @@ def update_lyft_infos(pkl_path, out_dir):
             empty_instance = get_empty_instance()
             empty_instance['bbox_3d'] = ori_info_dict['gt_boxes'][
                 i, :].tolist()
-            if ori_info_dict['gt_names'][i] in METAINFO['CLASSES']:
-                empty_instance['bbox_label'] = METAINFO['CLASSES'].index(
+            if ori_info_dict['gt_names'][i] in METAINFO['classes']:
+                empty_instance['bbox_label'] = METAINFO['classes'].index(
                     ori_info_dict['gt_names'][i])
             else:
                 ignore_class_name.add(ori_info_dict['gt_names'][i])
@@ -793,11 +832,20 @@ def update_lyft_infos(pkl_path, out_dir):
             temp_data_info['instances'].append(empty_instance)
         temp_data_info, _ = clear_data_info_unused_keys(temp_data_info)
         converted_list.append(temp_data_info)
-    pkl_name = pkl_path.split('/')[-1]
+    pkl_name = Path(pkl_path).name
     out_path = osp.join(out_dir, pkl_name)
     print(f'Writing to output file: {out_path}.')
     print(f'ignore classes: {ignore_class_name}')
-    converted_data_info = dict(metainfo=METAINFO, data_list=converted_list)
+
+    metainfo = dict()
+    metainfo['categories'] = {k: i for i, k in enumerate(METAINFO['classes'])}
+    if ignore_class_name:
+        for ignore_class in ignore_class_name:
+            metainfo['categories'][ignore_class] = -1
+    metainfo['dataset'] = 'lyft'
+    metainfo['version'] = data_list['metadata']['version']
+    metainfo['info_version'] = '1.1'
+    converted_data_info = dict(metainfo=metainfo, data_list=converted_list)
 
     mmengine.dump(converted_data_info, out_path, 'pkl')
 
@@ -820,7 +868,7 @@ def update_waymo_infos(pkl_path, out_dir):
     # TODO update to full label
     # TODO discuss how to process 'Van', 'DontCare'
     METAINFO = {
-        'CLASSES': ('Car', 'Pedestrian', 'Cyclist', 'Sign'),
+        'classes': ('Car', 'Pedestrian', 'Cyclist', 'Sign'),
     }
     print(f'Reading from input file: {pkl_path}.')
     data_list = mmengine.load(pkl_path)
@@ -851,7 +899,7 @@ def update_waymo_infos(pkl_path, out_dir):
                 ori_info_dict['calib'][f'P{cam_idx}'] @ lidar2cam).tolist()
 
         # image path
-        base_img_path = ori_info_dict['image']['image_path'].split('/')[-1]
+        base_img_path = Path(ori_info_dict['image']['image_path']).name
 
         for cam_idx, cam_key in enumerate(camera_types):
             temp_data_info['images'][cam_key]['timestamp'] = ori_info_dict[
@@ -867,8 +915,8 @@ def update_waymo_infos(pkl_path, out_dir):
             'point_cloud']['num_features']
         temp_data_info['lidar_points']['timestamp'] = ori_info_dict[
             'timestamp']
-        temp_data_info['lidar_points']['lidar_path'] = ori_info_dict[
-            'point_cloud']['velodyne_path'].split('/')[-1]
+        temp_data_info['lidar_points']['lidar_path'] = Path(
+            ori_info_dict['point_cloud']['velodyne_path']).name
 
         # TODO discuss the usage of Tr_velo_to_cam in lidar
         Trv2c = ori_info_dict['calib']['Tr_velo_to_cam'].astype(np.float32)
@@ -888,13 +936,13 @@ def update_waymo_infos(pkl_path, out_dir):
             lidar_sweep = get_single_lidar_sweep()
             lidar_sweep['ego2global'] = ori_sweep['pose']
             lidar_sweep['timestamp'] = ori_sweep['timestamp']
-            lidar_sweep['lidar_points']['lidar_path'] = ori_sweep[
-                'velodyne_path'].split('/')[-1]
+            lidar_sweep['lidar_points']['lidar_path'] = Path(
+                ori_sweep['velodyne_path']).name
             # image sweeps
             image_sweep = get_single_image_sweep(camera_types)
             image_sweep['ego2global'] = ori_sweep['pose']
             image_sweep['timestamp'] = ori_sweep['timestamp']
-            img_path = ori_sweep['image_path'].split('/')[-1]
+            img_path = Path(ori_sweep['image_path']).name
             for cam_idx, cam_key in enumerate(camera_types):
                 image_sweep['images'][cam_key]['img_path'] = img_path
 
@@ -910,8 +958,8 @@ def update_waymo_infos(pkl_path, out_dir):
             empty_instance = get_empty_instance()
             empty_instance['bbox'] = anns['bbox'][instance_id].tolist()
 
-            if anns['name'][instance_id] in METAINFO['CLASSES']:
-                empty_instance['bbox_label'] = METAINFO['CLASSES'].index(
+            if anns['name'][instance_id] in METAINFO['classes']:
+                empty_instance['bbox_label'] = METAINFO['classes'].index(
                     anns['name'][instance_id])
             else:
                 ignore_class_name.add(anns['name'][instance_id])
@@ -954,8 +1002,8 @@ def update_waymo_infos(pkl_path, out_dir):
             empty_instance = get_empty_instance()
             empty_instance['bbox'] = anns['bbox'][instance_id].tolist()
 
-            if anns['name'][instance_id] in METAINFO['CLASSES']:
-                empty_instance['bbox_label'] = METAINFO['CLASSES'].index(
+            if anns['name'][instance_id] in METAINFO['classes']:
+                empty_instance['bbox_label'] = METAINFO['classes'].index(
                     anns['name'][instance_id])
             else:
                 ignore_class_name.add(anns['name'][instance_id])
@@ -991,12 +1039,22 @@ def update_waymo_infos(pkl_path, out_dir):
 
         temp_data_info, _ = clear_data_info_unused_keys(temp_data_info)
         converted_list.append(temp_data_info)
-    pkl_name = pkl_path.split('/')[-1]
+    pkl_name = Path(pkl_path).name
     out_path = osp.join(out_dir, pkl_name)
     print(f'Writing to output file: {out_path}.')
     print(f'ignore classes: {ignore_class_name}')
-    converted_data_info = dict(
-        metainfo={'DATASET': 'Waymo'}, data_list=converted_list)
+
+    # dataset metainfo
+    metainfo = dict()
+    metainfo['categories'] = {k: i for i, k in enumerate(METAINFO['classes'])}
+    if ignore_class_name:
+        for ignore_class in ignore_class_name:
+            metainfo['categories'][ignore_class] = -1
+    metainfo['dataset'] = 'waymo'
+    metainfo['version'] = '1.2'
+    metainfo['info_version'] = '1.1'
+
+    converted_data_info = dict(metainfo=metainfo, data_list=converted_list)
 
     mmengine.dump(converted_data_info, out_path, 'pkl')
 
@@ -1073,4 +1131,4 @@ if __name__ == '__main__':
     if args.out_dir is None:
         args.out_dir = args.root_dir
     update_pkl_infos(
-        dataset=args.dataset, out_dir=args.out_dir, pkl_path=args.pkl)
+        dataset=args.dataset, out_dir=args.out_dir, pkl_path=args.pkl_path)

tools/dataset_converters/waymo_converter.py

@@ -12,7 +12,6 @@ except ImportError:
 from glob import glob
 from os.path import join
 
-import mmcv
 import mmengine
 import numpy as np
 import tensorflow as tf
@@ -70,10 +69,7 @@ class Waymo2KITTI(object):
             '_SIDE_LEFT',
             '_SIDE_RIGHT',
         ]
-        self.lidar_list = [
-            '_FRONT', '_FRONT_RIGHT', '_FRONT_LEFT', '_SIDE_RIGHT',
-            '_SIDE_LEFT'
-        ]
+        self.lidar_list = ['TOP', 'FRONT', 'SIDE_LEFT', 'SIDE_RIGHT', 'REAR']
         self.type_list = [
             'UNKNOWN', 'VEHICLE', 'PEDESTRIAN', 'SIGN', 'CYCLIST'
         ]
@@ -136,7 +132,9 @@ class Waymo2KITTI(object):
 
             self.save_image(frame, file_idx, frame_idx)
             self.save_calib(frame, file_idx, frame_idx)
-            self.save_lidar(frame, file_idx, frame_idx)
+            if 'testing_3d_camera_only_detection' not in self.load_dir:
+                # the camera only split doesn't contain lidar points.
+                self.save_lidar(frame, file_idx, frame_idx)
             self.save_pose(frame, file_idx, frame_idx)
             self.save_timestamp(frame, file_idx, frame_idx)
 
@@ -162,8 +160,8 @@ class Waymo2KITTI(object):
             img_path = f'{self.image_save_dir}{str(img.name - 1)}/' + \
                 f'{self.prefix}{str(file_idx).zfill(3)}' + \
                 f'{str(frame_idx).zfill(3)}.jpg'
-            img = mmcv.imfrombytes(img.image)
-            mmcv.imwrite(img, img_path)
+            with open(img_path, 'wb') as fp:
+                fp.write(img.image)
 
     def save_calib(self, frame, file_idx, frame_idx):
         """Parse and save the calibration data.
@@ -442,7 +440,6 @@ class Waymo2KITTI(object):
             dir_list1 = [
                 self.label_all_save_dir,
                 self.calib_save_dir,
-                self.point_cloud_save_dir,
                 self.pose_save_dir,
                 self.timestamp_save_dir,
             ]
@@ -452,10 +449,12 @@ class Waymo2KITTI(object):
                 dir_list2.append(self.cam_sync_label_save_dir)
         else:
             dir_list1 = [
-                self.calib_save_dir, self.point_cloud_save_dir,
-                self.pose_save_dir, self.timestamp_save_dir
+                self.calib_save_dir, self.pose_save_dir,
+                self.timestamp_save_dir
             ]
             dir_list2 = [self.image_save_dir]
+        if 'testing_3d_camera_only_detection' not in self.load_dir:
+            dir_list1.append(self.point_cloud_save_dir)
         for d in dir_list1:
             mmengine.mkdir_or_exist(d)
         for d in dir_list2:

tools/misc/browse_dataset.py

@@ -71,7 +71,7 @@ def build_data_cfg(config_path, aug, cfg_options):
     if aug:
         show_pipeline = cfg.train_pipeline
     else:
-        show_pipeline = cfg.eval_pipeline
+        show_pipeline = cfg.test_pipeline
         for i in range(len(cfg.train_pipeline)):
             if cfg.train_pipeline[i]['type'] == 'LoadAnnotations3D':
                 show_pipeline.insert(i, cfg.train_pipeline[i])
@@ -117,13 +117,20 @@ def main():
 
     progress_bar = ProgressBar(len(dataset))
 
-    for item in dataset:
+    for i, item in enumerate(dataset):
         # the 3D Boxes in input could be in any of three coordinates
         data_input = item['inputs']
         data_sample = item['data_samples'].numpy()
 
         out_file = osp.join(
-            args.output_dir) if args.output_dir is not None else None
+            args.output_dir,
+            f'{i}.jpg') if args.output_dir is not None else None
+
+        # o3d_save_path is valid when args.not_show is False
+        o3d_save_path = osp.join(args.output_dir, f'pc_{i}.png') if (
+            args.output_dir is not None
+            and vis_task in ['lidar_det', 'lidar_seg', 'multi-modality_det']
+            and not args.not_show) else None
 
         visualizer.add_datasample(
             '3d visualzier',
@@ -132,6 +139,7 @@ def main():
             show=not args.not_show,
             wait_time=args.show_interval,
             out_file=out_file,
+            o3d_save_path=o3d_save_path,
             vis_task=vis_task)
 
         progress_bar.update()