Release v1.0.0rc2

mmdet3d/models/losses/paconv_regularization_loss.py

@@ -3,8 +3,8 @@ import torch
 from torch import nn as nn
 
 from mmdet3d.ops import PAConv, PAConvCUDA
-from mmdet.models.builder import LOSSES
 from mmdet.models.losses.utils import weight_reduce_loss
+from ..builder import LOSSES
 
 
 def weight_correlation(conv):

mmdet3d/models/losses/uncertain_smooth_l1_loss.py

@@ -2,8 +2,8 @@
 import torch
 from torch import nn as nn
 
-from mmdet.models.builder import LOSSES
 from mmdet.models.losses.utils import weighted_loss
+from ..builder import LOSSES
 
 
 @weighted_loss

mmdet3d/models/middle_encoders/sparse_encoder.py

 # Copyright (c) OpenMMLab. All rights reserved.
-from mmcv.ops import SparseConvTensor, SparseSequential
+
 from mmcv.runner import auto_fp16
 from torch import nn as nn
 
 from mmdet3d.ops import SparseBasicBlock, make_sparse_convmodule
+from mmdet3d.ops.spconv import IS_SPCONV2_AVAILABLE
 from ..builder import MIDDLE_ENCODERS
 
+if IS_SPCONV2_AVAILABLE:
+    from spconv.pytorch import SparseConvTensor, SparseSequential
+else:
+    from mmcv.ops import SparseConvTensor, SparseSequential
+
 
 @MIDDLE_ENCODERS.register_module()
 class SparseEncoder(nn.Module):

mmdet3d/models/middle_encoders/sparse_unet.py

 # Copyright (c) OpenMMLab. All rights reserved.
 import torch
-from mmcv.ops import SparseConvTensor, SparseSequential
+
+from mmdet3d.ops.spconv import IS_SPCONV2_AVAILABLE
+
+if IS_SPCONV2_AVAILABLE:
+    from spconv.pytorch import SparseConvTensor, SparseSequential
+else:
+    from mmcv.ops import SparseConvTensor, SparseSequential
+
 from mmcv.runner import BaseModule, auto_fp16
 
 from mmdet3d.ops import SparseBasicBlock, make_sparse_convmodule

mmdet3d/models/necks/dla_neck.py

@@ -6,7 +6,7 @@ from mmcv.cnn import ConvModule, build_conv_layer
 from mmcv.runner import BaseModule
 from torch import nn as nn
 
-from mmdet.models.builder import NECKS
+from ..builder import NECKS
 
 
 def fill_up_weights(up):

mmdet3d/models/necks/imvoxel_neck.py

@@ -2,7 +2,7 @@
 from mmcv.cnn import ConvModule
 from torch import nn
 
-from mmdet.models import NECKS
+from ..builder import NECKS
 
 
 @NECKS.register_module()

mmdet3d/models/necks/pointnet2_fp_neck.py

@@ -3,7 +3,7 @@ from mmcv.runner import BaseModule
 from torch import nn as nn
 
 from mmdet3d.ops import PointFPModule
-from mmdet.models import NECKS
+from ..builder import NECKS
 
 
 @NECKS.register_module()

mmdet3d/models/necks/second_fpn.py

@@ -5,7 +5,7 @@ from mmcv.cnn import build_conv_layer, build_norm_layer, build_upsample_layer
 from mmcv.runner import BaseModule, auto_fp16
 from torch import nn as nn
 
-from mmdet.models import NECKS
+from ..builder import NECKS
 
 
 @NECKS.register_module()

mmdet3d/models/roi_heads/bbox_heads/h3d_bbox_head.py

@@ -7,11 +7,10 @@ from torch.nn import functional as F
 
 from mmdet3d.core.bbox import DepthInstance3DBoxes
 from mmdet3d.core.post_processing import aligned_3d_nms
-from mmdet3d.models.builder import build_loss
+from mmdet3d.models.builder import HEADS, build_loss
 from mmdet3d.models.losses import chamfer_distance
 from mmdet3d.ops import build_sa_module
 from mmdet.core import build_bbox_coder, multi_apply
-from mmdet.models import HEADS
 
 
 @HEADS.register_module()

mmdet3d/models/roi_heads/bbox_heads/parta2_bbox_head.py

@@ -3,17 +3,15 @@ import numpy as np
 import torch
 from mmcv.cnn import ConvModule, normal_init
 from mmcv.ops import SparseConvTensor, SparseMaxPool3d, SparseSequential
-from mmcv.ops import nms_bev as nms_gpu
-from mmcv.ops import nms_normal_bev as nms_normal_gpu
 from mmcv.runner import BaseModule
 from torch import nn as nn
 
 from mmdet3d.core.bbox.structures import (LiDARInstance3DBoxes,
                                           rotation_3d_in_axis, xywhr2xyxyr)
-from mmdet3d.models.builder import build_loss
+from mmdet3d.core.post_processing import nms_bev, nms_normal_bev
+from mmdet3d.models.builder import HEADS, build_loss
 from mmdet3d.ops import make_sparse_convmodule
 from mmdet.core import build_bbox_coder, multi_apply
-from mmdet.models import HEADS
 
 
 @HEADS.register_module()
@@ -582,9 +580,9 @@ class PartA2BboxHead(BaseModule):
             torch.Tensor: Selected indices.
         """
         if use_rotate_nms:
-            nms_func = nms_gpu
+            nms_func = nms_bev
         else:
-            nms_func = nms_normal_gpu
+            nms_func = nms_normal_bev
 
         assert box_probs.shape[
             1] == self.num_classes, f'box_probs shape: {str(box_probs.shape)}'

mmdet3d/models/roi_heads/bbox_heads/point_rcnn_bbox_head.py

@@ -3,17 +3,15 @@ import numpy as np
 import torch
 from mmcv.cnn import ConvModule, normal_init
 from mmcv.cnn.bricks import build_conv_layer
-from mmcv.ops import nms_bev as nms_gpu
-from mmcv.ops import nms_normal_bev as nms_normal_gpu
 from mmcv.runner import BaseModule
 from torch import nn as nn
 
 from mmdet3d.core.bbox.structures import (LiDARInstance3DBoxes,
                                           rotation_3d_in_axis, xywhr2xyxyr)
-from mmdet3d.models.builder import build_loss
+from mmdet3d.core.post_processing import nms_bev, nms_normal_bev
+from mmdet3d.models.builder import HEADS, build_loss
 from mmdet3d.ops import build_sa_module
 from mmdet.core import build_bbox_coder, multi_apply
-from mmdet.models import HEADS
 
 
 @HEADS.register_module()
@@ -239,7 +237,7 @@ class PointRCNNBboxHead(BaseModule):
         rcnn_reg = self.conv_reg(x_reg)
         rcnn_cls = rcnn_cls.transpose(1, 2).contiguous().squeeze(dim=1)
         rcnn_reg = rcnn_reg.transpose(1, 2).contiguous().squeeze(dim=1)
-        return (rcnn_cls, rcnn_reg)
+        return rcnn_cls, rcnn_reg
 
     def loss(self, cls_score, bbox_pred, rois, labels, bbox_targets,
              pos_gt_bboxes, reg_mask, label_weights, bbox_weights):
@@ -483,7 +481,7 @@ class PointRCNNBboxHead(BaseModule):
         local_roi_boxes[..., 0:3] = 0
         rcnn_boxes3d = self.bbox_coder.decode(local_roi_boxes, bbox_pred)
         rcnn_boxes3d[..., 0:3] = rotation_3d_in_axis(
-            rcnn_boxes3d[..., 0:3].unsqueeze(1), (roi_ry), axis=2).squeeze(1)
+            rcnn_boxes3d[..., 0:3].unsqueeze(1), roi_ry, axis=2).squeeze(1)
         rcnn_boxes3d[:, 0:3] += roi_xyz
 
         # post processing
@@ -492,7 +490,6 @@ class PointRCNNBboxHead(BaseModule):
             cur_class_labels = class_labels[batch_id]
             cur_cls_score = cls_score[roi_batch_id == batch_id].view(-1)
 
-            cur_box_prob = cls_score[batch_id]
             cur_box_prob = cur_cls_score.unsqueeze(1)
             cur_rcnn_boxes3d = rcnn_boxes3d[roi_batch_id == batch_id]
             keep = self.multi_class_nms(cur_box_prob, cur_rcnn_boxes3d,
@@ -524,7 +521,7 @@ class PointRCNNBboxHead(BaseModule):
             merging these two functions in the future.
 
         Args:
-            box_probs (torch.Tensor): Predicted boxes probabitilies in
+            box_probs (torch.Tensor): Predicted boxes probabilities in
                 shape (N,).
             box_preds (torch.Tensor): Predicted boxes in shape (N, 7+C).
             score_thr (float): Threshold of scores.
@@ -537,9 +534,9 @@ class PointRCNNBboxHead(BaseModule):
             torch.Tensor: Selected indices.
         """
         if use_rotate_nms:
-            nms_func = nms_gpu
+            nms_func = nms_bev
         else:
-            nms_func = nms_normal_gpu
+            nms_func = nms_normal_bev
 
         assert box_probs.shape[
             1] == self.num_classes, f'box_probs shape: {str(box_probs.shape)}'

mmdet3d/models/roi_heads/h3d_roi_head.py

 # Copyright (c) OpenMMLab. All rights reserved.
 from mmdet3d.core.bbox import bbox3d2result
-from mmdet.models import HEADS
-from ..builder import build_head
+from ..builder import HEADS, build_head
 from .base_3droi_head import Base3DRoIHead
 
 

mmdet3d/models/roi_heads/mask_heads/pointwise_semantic_head.py

@@ -5,9 +5,8 @@ from torch import nn as nn
 from torch.nn import functional as F
 
 from mmdet3d.core.bbox.structures import rotation_3d_in_axis
-from mmdet3d.models.builder import build_loss
+from mmdet3d.models.builder import HEADS, build_loss
 from mmdet.core import multi_apply
-from mmdet.models import HEADS
 
 
 @HEADS.register_module()

mmdet3d/models/roi_heads/mask_heads/primitive_head.py

@@ -6,11 +6,10 @@ from mmcv.runner import BaseModule
 from torch import nn as nn
 from torch.nn import functional as F
 
-from mmdet3d.models.builder import build_loss
+from mmdet3d.models.builder import HEADS, build_loss
 from mmdet3d.models.model_utils import VoteModule
 from mmdet3d.ops import build_sa_module
 from mmdet.core import multi_apply
-from mmdet.models import HEADS
 
 
 @HEADS.register_module()

mmdet3d/models/roi_heads/part_aggregation_roi_head.py

@@ -6,8 +6,7 @@ from torch.nn import functional as F
 from mmdet3d.core import AssignResult
 from mmdet3d.core.bbox import bbox3d2result, bbox3d2roi
 from mmdet.core import build_assigner, build_sampler
-from mmdet.models import HEADS
-from ..builder import build_head, build_roi_extractor
+from ..builder import HEADS, build_head, build_roi_extractor
 from .base_3droi_head import Base3DRoIHead
 
 

mmdet3d/models/roi_heads/point_rcnn_roi_head.py

@@ -5,8 +5,7 @@ from torch.nn import functional as F
 from mmdet3d.core import AssignResult
 from mmdet3d.core.bbox import bbox3d2result, bbox3d2roi
 from mmdet.core import build_assigner, build_sampler
-from mmdet.models import HEADS
-from ..builder import build_head, build_roi_extractor
+from ..builder import HEADS, build_head, build_roi_extractor
 from .base_3droi_head import Base3DRoIHead
 
 

mmdet3d/models/roi_heads/roi_extractors/single_roiaware_extractor.py

@@ -3,7 +3,7 @@ import torch
 from mmcv import ops
 from mmcv.runner import BaseModule
 
-from mmdet.models.builder import ROI_EXTRACTORS
+from mmdet3d.models.builder import ROI_EXTRACTORS
 
 
 @ROI_EXTRACTORS.register_module()

mmdet3d/models/roi_heads/roi_extractors/single_roipoint_extractor.py

@@ -4,7 +4,7 @@ from mmcv import ops
 from torch import nn as nn
 
 from mmdet3d.core.bbox.structures import rotation_3d_in_axis
-from mmdet.models.builder import ROI_EXTRACTORS
+from mmdet3d.models.builder import ROI_EXTRACTORS
 
 
 @ROI_EXTRACTORS.register_module()

mmdet3d/models/segmentors/encoder_decoder.py

@@ -5,8 +5,8 @@ from torch import nn as nn
 from torch.nn import functional as F
 
 from mmseg.core import add_prefix
-from mmseg.models import SEGMENTORS
-from ..builder import build_backbone, build_head, build_loss, build_neck
+from ..builder import (SEGMENTORS, build_backbone, build_head, build_loss,
+                       build_neck)
 from .base import Base3DSegmentor
 
 

mmdet3d/ops/norm.py

@@ -53,11 +53,27 @@ class NaiveSyncBatchNorm1d(nn.BatchNorm1d):
     # TODO: make mmcv fp16 utils handle customized norm layers
     @force_fp32(out_fp16=True)
     def forward(self, input):
+        """
+        Args:
+            input (tensor): Has shape (N, C) or (N, C, L), where N is
+                the batch size, C is the number of features or
+                channels, and L is the sequence length
+
+        Returns:
+            tensor: Has shape (N, C) or (N, C, L), has same shape
+            as input.
+        """
         assert input.dtype == torch.float32, \
             f'input should be in float32 type, got {input.dtype}'
-        if dist.get_world_size() == 1 or not self.training:
+        using_dist = dist.is_available() and dist.is_initialized()
+        if (not using_dist) or dist.get_world_size() == 1 \
+                or not self.training:
             return super().forward(input)
         assert input.shape[0] > 0, 'SyncBN does not support empty inputs'
+        is_two_dim = input.dim() == 2
+        if is_two_dim:
+            input = input.unsqueeze(2)
+
         C = input.shape[1]
         mean = torch.mean(input, dim=[0, 2])
         meansqr = torch.mean(input * input, dim=[0, 2])
@@ -76,7 +92,10 @@ class NaiveSyncBatchNorm1d(nn.BatchNorm1d):
         bias = self.bias - mean * scale
         scale = scale.reshape(1, -1, 1)
         bias = bias.reshape(1, -1, 1)
-        return input * scale + bias
+        output = input * scale + bias
+        if is_two_dim:
+            output = output.squeeze(2)
+        return output
 
 
 @NORM_LAYERS.register_module('naiveSyncBN2d')
@@ -107,9 +126,19 @@ class NaiveSyncBatchNorm2d(nn.BatchNorm2d):
     # TODO: make mmcv fp16 utils handle customized norm layers
     @force_fp32(out_fp16=True)
     def forward(self, input):
+        """
+        Args:
+            Input (tensor): Feature has shape (N, C, H, W).
+
+        Returns:
+            tensor: Has shape (N, C, H, W), same shape as input.
+        """
         assert input.dtype == torch.float32, \
             f'input should be in float32 type, got {input.dtype}'
-        if dist.get_world_size() == 1 or not self.training:
+        using_dist = dist.is_available() and dist.is_initialized()
+        if (not using_dist) or \
+                dist.get_world_size() == 1 or \
+                not self.training:
             return super().forward(input)
 
         assert input.shape[0] > 0, 'SyncBN does not support empty inputs'