[Fix] Fix unittest with GPU

mmdet3d/models/dense_heads/point_rpn_head.py

@@ -3,6 +3,7 @@
 from mmcv.runner import BaseModule, force_fp32
 from torch import nn as nn
 
+from mmdet3d.core import xywhr2xyxyr
 from mmdet3d.core.bbox.structures import (DepthInstance3DBoxes,
                                           LiDARInstance3DBoxes)
 from mmdet3d.ops.iou3d.iou3d_utils import nms_gpu, nms_normal_gpu
@@ -321,26 +322,31 @@ def class_agnostic_nms(self, obj_scores, sem_scores, bbox, points,
         else:
             raise NotImplementedError('Unsupported bbox type!')
 
-        bbox = bbox.tensor[nonempty_box_mask]
+        obj_scores = obj_scores[nonempty_box_mask]
+        sem_scores = sem_scores[nonempty_box_mask]
+        bbox_selected = bbox.tensor[nonempty_box_mask]
+        bbox_for_nms = xywhr2xyxyr(bbox.bev)[nonempty_box_mask]
 
         if self.test_cfg.score_thr is not None:
             score_thr = self.test_cfg.score_thr
             keep = (obj_scores >= score_thr)
             obj_scores = obj_scores[keep]
             sem_scores = sem_scores[keep]
-            bbox = bbox[keep]
+            bbox_selected = bbox_selected[keep]
+            bbox_for_nms = bbox_for_nms[keep]
 
         if obj_scores.shape[0] > 0:
             topk = min(nms_cfg.nms_pre, obj_scores.shape[0])
             obj_scores_nms, indices = torch.topk(obj_scores, k=topk)
-            bbox_for_nms = bbox[indices]
+            bbox_selected = bbox_selected[indices]
+            bbox_for_nms = bbox_for_nms[indices]
             sem_scores_nms = sem_scores[indices]
 
             keep = nms_func(bbox_for_nms[:, 0:7], obj_scores_nms,
                             nms_cfg.iou_thr)
             keep = keep[:nms_cfg.nms_post]
 
-            bbox_selected = bbox_for_nms[keep]
+            bbox_selected = bbox_selected[keep]
             score_selected = obj_scores_nms[keep]
             cls_preds = sem_scores_nms[keep]
             labels = torch.argmax(cls_preds, -1)

tests/test_data/test_datasets/test_kitti_dataset.py

@@ -250,9 +250,12 @@ def test_evaluate():
     metric = ['mAP']
     result = dict(boxes_3d=boxes_3d, labels_3d=labels_3d, scores_3d=scores_3d)
     ap_dict = kitti_dataset.evaluate([result], metric)
-    assert np.isclose(ap_dict['KITTI/Overall_3D_easy'], 3.0303030303030307)
-    assert np.isclose(ap_dict['KITTI/Overall_3D_moderate'], 3.0303030303030307)
-    assert np.isclose(ap_dict['KITTI/Overall_3D_hard'], 3.0303030303030307)
+    assert np.isclose(ap_dict['KITTI/Overall_3D_AP11_easy'],
+                      3.0303030303030307)
+    assert np.isclose(ap_dict['KITTI/Overall_3D_AP11_moderate'],
+                      3.0303030303030307)
+    assert np.isclose(ap_dict['KITTI/Overall_3D_AP11_hard'],
+                      3.0303030303030307)
 
 
 def test_show():

tests/test_data/test_datasets/test_kitti_mono_dataset.py

@@ -207,9 +207,11 @@ def test_evaluate():
 
     metric = ['mAP']
     ap_dict = kitti_dataset.evaluate(results, metric)
-    assert np.isclose(ap_dict['img_bbox/KITTI/Overall_3D_easy'], 3.0303)
-    assert np.isclose(ap_dict['img_bbox/KITTI/Overall_3D_moderate'], 6.0606)
-    assert np.isclose(ap_dict['img_bbox/KITTI/Overall_3D_hard'], 6.0606)
-    assert np.isclose(ap_dict['img_bbox2d/KITTI/Overall_2D_easy'], 3.0303)
-    assert np.isclose(ap_dict['img_bbox2d/KITTI/Overall_2D_moderate'], 6.0606)
-    assert np.isclose(ap_dict['img_bbox2d/KITTI/Overall_2D_hard'], 6.0606)
+    assert np.isclose(ap_dict['img_bbox/KITTI/Overall_3D_AP11_easy'], 3.0303)
+    assert np.isclose(ap_dict['img_bbox/KITTI/Overall_3D_AP11_moderate'],
+                      6.0606)
+    assert np.isclose(ap_dict['img_bbox/KITTI/Overall_3D_AP11_hard'], 6.0606)
+    assert np.isclose(ap_dict['img_bbox2d/KITTI/Overall_2D_AP11_easy'], 3.0303)
+    assert np.isclose(ap_dict['img_bbox2d/KITTI/Overall_2D_AP11_moderate'],
+                      6.0606)
+    assert np.isclose(ap_dict['img_bbox2d/KITTI/Overall_2D_AP11_hard'], 6.0606)

tests/test_data/test_datasets/test_waymo_dataset.py

@@ -144,9 +144,12 @@ def test_evaluate():
     # kitti protocol
     metric = ['kitti']
     ap_dict = waymo_dataset.evaluate([result], metric=metric)
-    assert np.isclose(ap_dict['KITTI/Overall_3D_easy'], 3.0303030303030307)
-    assert np.isclose(ap_dict['KITTI/Overall_3D_moderate'], 3.0303030303030307)
-    assert np.isclose(ap_dict['KITTI/Overall_3D_hard'], 3.0303030303030307)
+    assert np.isclose(ap_dict['KITTI/Overall_3D_AP11_easy'],
+                      3.0303030303030307)
+    assert np.isclose(ap_dict['KITTI/Overall_3D_AP11_moderate'],
+                      3.0303030303030307)
+    assert np.isclose(ap_dict['KITTI/Overall_3D_AP11_hard'],
+                      3.0303030303030307)
 
     # waymo protocol
     metric = ['waymo']

tests/test_models/test_common_modules/test_dgcnn_modules.py

@@ -12,11 +12,11 @@ def test_dgcnn_gf_module():
     self = DGCNNGFModule(
         mlp_channels=[18, 64, 64],
         num_sample=20,
-        knn_mod='D-KNN',
+        knn_mode='D-KNN',
         radius=None,
         norm_cfg=dict(type='BN2d'),
         act_cfg=dict(type='ReLU'),
-        pool_mod='max').cuda()
+        pool_mode='max').cuda()
 
     assert self.mlps[0].layer0.conv.in_channels == 18
     assert self.mlps[0].layer0.conv.out_channels == 64
@@ -36,11 +36,11 @@ def test_dgcnn_gf_module():
     self = DGCNNGFModule(
         mlp_channels=[6, 64, 64],
         num_sample=20,
-        knn_mod='F-KNN',
+        knn_mode='F-KNN',
         radius=None,
         norm_cfg=dict(type='BN2d'),
         act_cfg=dict(type='ReLU'),
-        pool_mod='max').cuda()
+        pool_mode='max').cuda()
 
     # test forward
     new_points = self(xyz)
@@ -50,11 +50,11 @@ def test_dgcnn_gf_module():
     self = DGCNNGFModule(
         mlp_channels=[6, 64, 64],
         num_sample=20,
-        knn_mod='F-KNN',
+        knn_mode='F-KNN',
         radius=0.2,
         norm_cfg=dict(type='BN2d'),
         act_cfg=dict(type='ReLU'),
-        pool_mod='max').cuda()
+        pool_mode='max').cuda()
 
 
 def test_dgcnn_fa_module():

tests/test_models/test_detectors.py

@@ -473,11 +473,11 @@ def test_imvoxelnet():
     assert labels_3d.shape[0] >= 0
 
 
-def test_pointrcnn():
+def test_point_rcnn():
     if not torch.cuda.is_available():
         pytest.skip('test requires GPU and torch+cuda')
     pointrcnn_cfg = _get_detector_cfg(
-        'pointrcnn/pointrcnn_2x8_kitti-3d-3classes.py')
+        'point_rcnn/point_rcnn_2x8_kitti-3d-3classes.py')
     self = build_detector(pointrcnn_cfg).cuda()
     points_0 = torch.rand([1000, 4], device='cuda')
     points_1 = torch.rand([1000, 4], device='cuda')
@@ -539,9 +539,8 @@ def test_smoke():
     attr_labels = None
     img_metas = [
         dict(
-            cam_intrinsic=[[721.5377, 0., 609.5593, 0.],
-                           [0., 721.5377, 172.854, 0.], [0., 0., 1., 0.],
-                           [0., 0., 0., 1.]],
+            cam2img=[[721.5377, 0., 609.5593, 0.], [0., 721.5377, 172.854, 0.],
+                     [0., 0., 1., 0.], [0., 0., 0., 1.]],
             scale_factor=np.array([1., 1., 1., 1.], dtype=np.float32),
             pad_shape=[384, 1280],
             trans_mat=np.array([[0.25, 0., 0.], [0., 0.25, 0], [0., 0., 1.]],

tests/test_models/test_heads/test_heads.py

@@ -52,7 +52,6 @@ def _get_head_cfg(fname):
     These are deep copied to allow for safe modification of parameters without
     influencing other tests.
     """
-    import mmcv
     config = _get_config_module(fname)
     model = copy.deepcopy(config.model)
     train_cfg = mmcv.Config(copy.deepcopy(config.model.train_cfg))
@@ -70,7 +69,6 @@ def _get_rpn_head_cfg(fname):
     These are deep copied to allow for safe modification of parameters without
     influencing other tests.
     """
-    import mmcv
     config = _get_config_module(fname)
     model = copy.deepcopy(config.model)
     train_cfg = mmcv.Config(copy.deepcopy(config.model.train_cfg))
@@ -88,7 +86,6 @@ def _get_roi_head_cfg(fname):
     These are deep copied to allow for safe modification of parameters without
     influencing other tests.
     """
-    import mmcv
     config = _get_config_module(fname)
     model = copy.deepcopy(config.model)
     train_cfg = mmcv.Config(copy.deepcopy(config.model.train_cfg))
@@ -106,7 +103,6 @@ def _get_pts_bbox_head_cfg(fname):
     These are deep copied to allow for safe modification of parameters without
     influencing other tests.
     """
-    import mmcv
     config = _get_config_module(fname)
     model = copy.deepcopy(config.model)
     train_cfg = mmcv.Config(copy.deepcopy(config.model.train_cfg.pts))
@@ -132,7 +128,7 @@ def _get_pointrcnn_rpn_head_cfg(fname):
     rpn_head = model.rpn_head
     rpn_head.update(train_cfg=train_cfg.rpn)
     rpn_head.update(test_cfg=test_cfg.rpn)
-    return rpn_head, train_cfg.rpn.rpn_proposal
+    return rpn_head, train_cfg.rpn
 
 
 def _get_vote_head_cfg(fname):
@@ -141,7 +137,6 @@ def _get_vote_head_cfg(fname):
     These are deep copied to allow for safe modification of parameters without
     influencing other tests.
     """
-    import mmcv
     config = _get_config_module(fname)
     model = copy.deepcopy(config.model)
     train_cfg = mmcv.Config(copy.deepcopy(config.model.train_cfg))
@@ -290,11 +285,11 @@ def test_parta2_rpnhead_getboxes():
     assert result_list[0]['boxes_3d'].tensor.shape == torch.Size([512, 7])
 
 
-def test_pointrcnn_rpnhead_getboxes():
+def test_point_rcnn_rpnhead_getboxes():
     if not torch.cuda.is_available():
         pytest.skip('test requires GPU and torch+cuda')
     rpn_head_cfg, proposal_cfg = _get_pointrcnn_rpn_head_cfg(
-        './pointrcnn/pointrcnn_2x8_kitti-3d-3classes.py')
+        './point_rcnn/point_rcnn_2x8_kitti-3d-3classes.py')
     self = build_head(rpn_head_cfg)
     self.cuda()
 
@@ -315,7 +310,7 @@ def test_pointrcnn_rpnhead_getboxes():
     assert cls_preds.shape == (2, 1024, 3)
     points = torch.rand([2, 1024, 3], dtype=torch.float32).cuda()
     result_list = self.get_bboxes(points, bbox_preds, cls_preds, input_metas)
-    max_num = proposal_cfg.max_num
+    max_num = proposal_cfg.nms_cfg.nms_post
     bbox, score_selected, labels, cls_preds_selected = result_list[0]
     assert bbox.tensor.shape == (max_num, 7)
     assert score_selected.shape == (max_num, )
@@ -515,22 +510,24 @@ def test_smoke_mono3d_head():
 
 
 def test_parta2_bbox_head():
+    if not torch.cuda.is_available():
+        pytest.skip('test requires GPU and torch+cuda')
     parta2_bbox_head_cfg = _get_parta2_bbox_head_cfg(
         './parta2/hv_PartA2_secfpn_2x8_cyclic_80e_kitti-3d-3class.py')
-    self = build_head(parta2_bbox_head_cfg)
-    seg_feats = torch.rand([256, 14, 14, 14, 16])
-    part_feats = torch.rand([256, 14, 14, 14, 4])
+    self = build_head(parta2_bbox_head_cfg).cuda()
+    seg_feats = torch.rand([256, 14, 14, 14, 16]).cuda()
+    part_feats = torch.rand([256, 14, 14, 14, 4]).cuda()
 
     cls_score, bbox_pred = self.forward(seg_feats, part_feats)
     assert cls_score.shape == (256, 1)
     assert bbox_pred.shape == (256, 7)
 
 
-def test_pointrcnn_bbox_head():
+def test_point_rcnn_bbox_head():
     if not torch.cuda.is_available():
         pytest.skip('test requires GPU and torch+cuda')
     pointrcnn_bbox_head_cfg = _get_pointrcnn_bbox_head_cfg(
-        './pointrcnn/pointrcnn_2x8_kitti-3d-3classes.py')
+        './point_rcnn/point_rcnn_2x8_kitti-3d-3classes.py')
     self = build_head(pointrcnn_bbox_head_cfg).cuda()
     feats = torch.rand([100, 512, 133]).cuda()
     rcnn_cls, rcnn_reg = self.forward(feats)
@@ -612,12 +609,12 @@ def test_part_aggregation_ROI_head():
     assert labels_3d.shape == (12, )
 
 
-def test_pointrcnn_roi_head():
+def test_point_rcnn_roi_head():
     if not torch.cuda.is_available():
         pytest.skip('test requires GPU and torch+cuda')
 
     roi_head_cfg = _get_roi_head_cfg(
-        './pointrcnn/pointrcnn_2x8_kitti-3d-3classes.py')
+        './point_rcnn/point_rcnn_2x8_kitti-3d-3classes.py')
 
     self = build_head(roi_head_cfg).cuda()
 

tests/test_models/test_heads/test_roi_extractors.py

@@ -37,8 +37,7 @@ def test_single_roipoint_extractor():
     if not torch.cuda.is_available():
         pytest.skip('test requires GPU and torch+cuda')
 
-    roi_layer_cfg = dict(
-        type='RoIPointPool3d', num_sampled_points=512, pool_extra_width=0)
+    roi_layer_cfg = dict(type='RoIPointPool3d', num_sampled_points=512)
 
     self = Single3DRoIPointExtractor(roi_layer=roi_layer_cfg)
 

tests/test_models/test_necks/test_necks.py

@@ -114,7 +114,7 @@ def test_dla_neck():
             for i in range(len(in_channels))
         ]
         outputs = neck(feats)
-        assert outputs.shape == (4, 64, 8, 8)
+        assert outputs[0].shape == (4, 64, 8, 8)
     else:
         # Test DLA Neck without DCNv2 on CPU
         neck_cfg = dict(