[Refactor] 3D Segmentor and EncoderDecoder3D

tests/test_models/test_decode_heads/test_paconv_head.py

+# Copyright (c) OpenMMLab. All rights reserved.
+from unittest import TestCase
+
+import torch
+
+from mmdet3d.core import Det3DDataSample, PointData
+from mmdet3d.models.decode_heads import PAConvHead
+
+
+class TestPAConvHead(TestCase):
+
+    def test_paconv_head_loss(self):
+        """Tests PAConv head loss."""
+
+        paconv_head = PAConvHead(
+            fp_channels=((768, 256, 256), (384, 256, 256), (320, 256, 128),
+                         (128 + 6, 128, 128, 128)),
+            channels=128,
+            num_classes=20,
+            dropout_ratio=0.5,
+            conv_cfg=dict(type='Conv1d'),
+            norm_cfg=dict(type='BN1d'),
+            act_cfg=dict(type='ReLU'),
+            loss_decode=dict(
+                type='mmdet.CrossEntropyLoss',
+                use_sigmoid=False,
+                class_weight=None,
+                loss_weight=1.0),
+            ignore_index=20)
+
+        # PAConv head expects dict format features
+        sa_xyz = [
+            torch.rand(1, 4096, 3).float(),
+            torch.rand(1, 1024, 3).float(),
+            torch.rand(1, 256, 3).float(),
+            torch.rand(1, 64, 3).float(),
+            torch.rand(1, 16, 3).float(),
+        ]
+        sa_features = [
+            torch.rand(1, 6, 4096).float(),
+            torch.rand(1, 64, 1024).float(),
+            torch.rand(1, 128, 256).float(),
+            torch.rand(1, 256, 64).float(),
+            torch.rand(1, 512, 16).float(),
+        ]
+        feat_dict = dict(sa_xyz=sa_xyz, sa_features=sa_features)
+
+        # Test forward
+        seg_logits = paconv_head.forward(feat_dict)
+
+        self.assertEqual(seg_logits, torch.Size([1, 20, 4096]))
+
+        # When truth is non-empty then losses
+        # should be nonzero for random inputs
+        pts_semantic_mask = torch.randint(0, 20, (4096)).long()
+        gt_pts_seg = PointData(pts_semantic_mask=pts_semantic_mask)
+
+        datasample = Det3DDataSample()
+        datasample.gt_pts_seg = gt_pts_seg
+
+        gt_losses = paconv_head.loss(seg_logits, [datasample])
+
+        gt_sem_seg_loss = gt_losses['loss_sem_seg'].item()
+
+        self.assertGreater(gt_sem_seg_loss, 0,
+                           'semantic seg loss should be positive')

tests/test_models/test_decode_heads/test_pointnet2_head.py

+# Copyright (c) OpenMMLab. All rights reserved.
+from unittest import TestCase
+
+import torch
+
+from mmdet3d.core import Det3DDataSample, PointData
+from mmdet3d.models.decode_heads import PointNet2Head
+
+
+class TestPointNet2Head(TestCase):
+
+    def test_paconv_head_loss(self):
+        """Tests PAConv head loss."""
+
+        pointnet2_head = PointNet2Head(
+            fp_channels=((768, 256, 256), (384, 256, 256), (320, 256, 128),
+                         (128, 128, 128, 128)),
+            channels=128,
+            num_classes=20,
+            dropout_ratio=0.5,
+            conv_cfg=dict(type='Conv1d'),
+            norm_cfg=dict(type='BN1d'),
+            act_cfg=dict(type='ReLU'),
+            loss_decode=dict(
+                type='mmdet.CrossEntropyLoss',
+                use_sigmoid=False,
+                class_weight=None,
+                loss_weight=1.0),
+            ignore_index=20)
+
+        # DGCNN head expects dict format features
+        sa_xyz = [
+            torch.rand(1, 4096, 3).float(),
+            torch.rand(1, 1024, 3).float(),
+            torch.rand(1, 256, 3).float(),
+            torch.rand(1, 64, 3).float(),
+            torch.rand(1, 16, 3).float(),
+        ]
+        sa_features = [
+            torch.rand(1, 6, 4096).float(),
+            torch.rand(1, 64, 1024).float(),
+            torch.rand(1, 128, 256).float(),
+            torch.rand(1, 256, 64).float(),
+            torch.rand(1, 512, 16).float(),
+        ]
+        feat_dict = dict(sa_xyz=sa_xyz, sa_features=sa_features)
+
+        # Test forward
+        seg_logits = pointnet2_head.forward(feat_dict)
+
+        self.assertEqual(seg_logits, torch.Size([1, 20, 4096]))
+
+        # When truth is non-empty then losses
+        # should be nonzero for random inputs
+        pts_semantic_mask = torch.randint(0, 20, (4096)).long()
+        gt_pts_seg = PointData(pts_semantic_mask=pts_semantic_mask)
+
+        datasample = Det3DDataSample()
+        datasample.gt_pts_seg = gt_pts_seg
+
+        gt_losses = pointnet2_head.loss(seg_logits, [datasample])
+
+        gt_sem_seg_loss = gt_losses['loss_sem_seg'].item()
+
+        self.assertGreater(gt_sem_seg_loss, 0,
+                           'semantic seg loss should be positive')