[Feature]Add Lidarseg benchmark (#2530)

* enhance minkunet

* add 2x

* add config

* add flip...

* add bottleneck

* add spvcnn & cylinder3d

* add spvcnn & cylinder3d

* refactor minkunet & spvcnn

* add minkv2

* fix mink34 shared res block

* add mink spconv

* fix spconv int32 max bug

* fix spconv int32 max bug2

* add minkowski backends

* rename config

* fix minkv2 config

* fix max voxel bug

* add checkpointhook mink18

* add backbone docstring

* fix torchsparse uninstall bug

* remove ME

* fix ut

* fix cylinder3d config

Author: sunjiahao1999

configs/_base_/models/minkunet.py

@@ -4,19 +4,24 @@
         type='Det3DDataPreprocessor',
         voxel=True,
         voxel_type='minkunet',
+        batch_first=False,
+        max_voxels=80000,
         voxel_layer=dict(
             max_num_points=-1,
             point_cloud_range=[-100, -100, -20, 100, 100, 20],
             voxel_size=[0.05, 0.05, 0.05],
-            max_voxels=(-1, -1)),
-        max_voxels=80000),
+            max_voxels=(-1, -1))),
     backbone=dict(
         type='MinkUNetBackbone',
         in_channels=4,
-        base_channels=32,
         num_stages=4,
+        base_channels=32,
         encoder_channels=[32, 64, 128, 256],
-        decoder_channels=[256, 128, 96, 96]),
+        encoder_blocks=[2, 2, 2, 2],
+        decoder_channels=[256, 128, 96, 96],
+        decoder_blocks=[2, 2, 2, 2],
+        block_type='basic',
+        sparseconv_backend='torchsparse'),
     decode_head=dict(
         type='MinkUNetHead',
         channels=96,

configs/_base_/models/spvcnn.py

@@ -4,19 +4,24 @@
         type='Det3DDataPreprocessor',
         voxel=True,
         voxel_type='minkunet',
+        batch_first=False,
+        max_voxels=80000,
         voxel_layer=dict(
             max_num_points=-1,
             point_cloud_range=[-100, -100, -20, 100, 100, 20],
             voxel_size=[0.05, 0.05, 0.05],
-            max_voxels=(-1, -1)),
-        max_voxels=80000),
+            max_voxels=(-1, -1))),
     backbone=dict(
         type='SPVCNNBackbone',
         in_channels=4,
-        base_channels=32,
         num_stages=4,
+        base_channels=32,
         encoder_channels=[32, 64, 128, 256],
+        encoder_blocks=[2, 2, 2, 2],
         decoder_channels=[256, 128, 96, 96],
+        decoder_blocks=[2, 2, 2, 2],
+        block_type='basic',
+        sparseconv_backend='torchsparse',
         drop_ratio=0.3),
     decode_head=dict(
         type='MinkUNetHead',

configs/_base_/schedules/schedule-3x.py

@@ -8,7 +8,7 @@
     clip_grad=dict(max_norm=10, norm_type=2),
 )
 
-# training schedule for 1x
+# training schedule for 3x
 train_cfg = dict(type='EpochBasedTrainLoop', max_epochs=36, val_interval=1)
 val_cfg = dict(type='ValLoop')
 test_cfg = dict(type='TestLoop')

configs/cylinder3d/cylinder3d_4xb4_3x_semantickitti.py renamed to configs/cylinder3d/cylinder3d_4xb4-3x_semantickitti.py

@@ -4,7 +4,6 @@
 ]
 
 # optimizer
-# This schedule is mainly used by models on nuScenes dataset
 lr = 0.001
 optim_wrapper = dict(
     type='OptimWrapper',

configs/cylinder3d/cylinder3d_8xb2-laser-polar-mix-3x_semantickitti.py

@@ -0,0 +1,80 @@
+_base_ = [
+    '../_base_/datasets/semantickitti.py', '../_base_/models/cylinder3d.py',
+    '../_base_/schedules/schedule-3x.py', '../_base_/default_runtime.py'
+]
+
+train_pipeline = [
+    dict(type='LoadPointsFromFile', coord_type='LIDAR', load_dim=4, use_dim=4),
+    dict(
+        type='LoadAnnotations3D',
+        with_bbox_3d=False,
+        with_label_3d=False,
+        with_seg_3d=True,
+        seg_3d_dtype='np.int32',
+        seg_offset=2**16,
+        dataset_type='semantickitti'),
+    dict(type='PointSegClassMapping'),
+    dict(
+        type='RandomChoice',
+        transforms=[
+            [
+                dict(
+                    type='LaserMix',
+                    num_areas=[3, 4, 5, 6],
+                    pitch_angles=[-25, 3],
+                    pre_transform=[
+                        dict(
+                            type='LoadPointsFromFile',
+                            coord_type='LIDAR',
+                            load_dim=4,
+                            use_dim=4),
+                        dict(
+                            type='LoadAnnotations3D',
+                            with_bbox_3d=False,
+                            with_label_3d=False,
+                            with_seg_3d=True,
+                            seg_3d_dtype='np.int32',
+                            seg_offset=2**16,
+                            dataset_type='semantickitti'),
+                        dict(type='PointSegClassMapping')
+                    ],
+                    prob=1)
+            ],
+            [
+                dict(
+                    type='PolarMix',
+                    instance_classes=[0, 1, 2, 3, 4, 5, 6, 7],
+                    swap_ratio=0.5,
+                    rotate_paste_ratio=1.0,
+                    pre_transform=[
+                        dict(
+                            type='LoadPointsFromFile',
+                            coord_type='LIDAR',
+                            load_dim=4,
+                            use_dim=4),
+                        dict(
+                            type='LoadAnnotations3D',
+                            with_bbox_3d=False,
+                            with_label_3d=False,
+                            with_seg_3d=True,
+                            seg_3d_dtype='np.int32',
+                            seg_offset=2**16,
+                            dataset_type='semantickitti'),
+                        dict(type='PointSegClassMapping')
+                    ],
+                    prob=1)
+            ],
+        ],
+        prob=[0.5, 0.5]),
+    dict(
+        type='GlobalRotScaleTrans',
+        rot_range=[0., 6.28318531],
+        scale_ratio_range=[0.95, 1.05],
+        translation_std=[0, 0, 0],
+    ),
+    dict(type='Pack3DDetInputs', keys=['points', 'pts_semantic_mask'])
+]
+
+train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
+
+default_hooks = dict(checkpoint=dict(type='CheckpointHook', interval=1))

configs/minkunet/minkunet_w16_8xb2-15e_semantickitti.py renamed to configs/minkunet/minkunet18_w16_torchsparse_8xb2-amp-15e_semantickitti.py

@@ -0,0 +1,7 @@
+_base_ = [
+    './minkunet34_w32_torchsparse_8xb2-laser-polar-mix-3x_semantickitti.py'
+]
+
+model = dict(
+    data_preprocessor=dict(batch_first=True),
+    backbone=dict(sparseconv_backend='minkowski'))

configs/minkunet/minkunet_w20_8xb2-15e_semantickitti.py renamed to configs/minkunet/minkunet18_w20_torchsparse_8xb2-amp-15e_semantickitti.py

@@ -0,0 +1,9 @@
+_base_ = [
+    './minkunet34_w32_torchsparse_8xb2-laser-polar-mix-3x_semantickitti.py'
+]
+
+model = dict(
+    data_preprocessor=dict(batch_first=True),
+    backbone=dict(sparseconv_backend='spconv'))
+
+optim_wrapper = dict(type='AmpOptimWrapper', loss_scale='dynamic')