[Fix] Fix some bugs to be compatible with the latest mmcv and mmengine (

#687) * fix some import errors about mmcv and unit test * update the version of torchvision * skip cpu ut for prdimp * fix UT about mask2former * fix prdimp ut * reduce reid channel in strongsort Co-authored-by: Pengxiang Li <[email protected]>
open-mmlab · Aug 24, 2022 · 067cc16 · 067cc16
1 parent 9208168
commit 067cc16
Show file tree

Hide file tree

Showing 34 changed files with 106 additions and 98 deletions.
diff --git a/.circleci/test.yml b/.circleci/test.yml
@@ -153,8 +153,8 @@ workflows:
                 - dev-1.x
       - build_cpu:
           name: minimum_version_cpu
-          torch: 1.6.0
-          torchvision: 0.7.0
+          torch: 1.7.0
+          torchvision: 0.8.1
           python: 3.6.9  # The lowest python 3.6.x version available on CircleCI images
           mmcv: https://download.openmmlab.com/mmcv/dev-2.x/cpu/torch1.6.0/mmcv_full-2.0.0rc0-cp36-cp36m-manylinux1_x86_64.whl
           requires:
@@ -187,7 +187,7 @@ workflows:
     jobs:
       - build_cuda:
           name: minimum_version_gpu
-          torch: 1.6.0
+          torch: 1.7.0
           # Use double quotation mark to explicitly specify its type
           # as string instead of number
           mmcv: https://download.openmmlab.com/mmcv/dev-2.x/cu101/torch1.6.0/mmcv_full-2.0.0rc0-cp37-cp37m-manylinux1_x86_64.whl

diff --git a/.dev_scripts/benchmark/gather_train_benchmark_metric.py b/.dev_scripts/benchmark/gather_train_benchmark_metric.py
@@ -4,7 +4,7 @@
 import json
 import os.path as osp
 
-import mmcv
+import mmengine
 
 try:
     import xlrd
@@ -67,7 +67,7 @@ def parse_args():
             if osp.exists(result_path):
 
                 # 1 read config and excel
-                cfg = mmcv.Config.fromfile(config)
+                cfg = mmengine.Config.fromfile(config)
                 total_epochs = cfg.train_cfg.max_epochs if cfg.train_cfg else 0
 
                 # the first metric will be used to find the best ckpt

diff --git a/configs/sot/stark/stark-st1_r50_8xb16-500e_got10k-lasot-trackingnet-coco_base.py b/configs/sot/stark/stark-st1_r50_8xb16-500e_got10k-lasot-trackingnet-coco_base.py
@@ -35,7 +35,7 @@
         transformer=dict(
             type='StarkTransformer',
             encoder=dict(
-                type='DetrTransformerEncoder',
+                type='mmdet.DetrTransformerEncoder',
                 num_layers=6,
                 transformerlayers=dict(
                     type='BaseTransformerLayer',
@@ -53,7 +53,7 @@
                         ffn_drop=0.1),
                     operation_order=('self_attn', 'norm', 'ffn', 'norm'))),
             decoder=dict(
-                type='DetrTransformerDecoder',
+                type='mmdet.DetrTransformerDecoder',
                 return_intermediate=False,
                 num_layers=6,
                 transformerlayers=dict(
@@ -72,7 +72,8 @@
                                      'ffn', 'norm'))),
         ),
         positional_encoding=dict(
-            type='SinePositionalEncoding', num_feats=128, normalize=True),
+            type='mmdet.SinePositionalEncoding', num_feats=128,
+            normalize=True),
         bbox_head=dict(
             type='CornerPredictorHead',
             inplanes=256,

diff --git a/configs/vis/mask2former/mask2former_r50_8xb2-8e_youtubevis2019.py b/configs/vis/mask2former/mask2former_r50_8xb2-8e_youtubevis2019.py
@@ -35,12 +35,12 @@
         num_frames=num_frames,
         num_transformer_feat_level=3,
         pixel_decoder=dict(
-            type='MSDeformAttnPixelDecoder',
+            type='mmdet.MSDeformAttnPixelDecoder',
             num_outs=3,
             norm_cfg=dict(type='GN', num_groups=32),
             act_cfg=dict(type='ReLU'),
             encoder=dict(
-                type='DetrTransformerEncoder',
+                type='mmdet.DetrTransformerEncoder',
                 num_layers=6,
                 transformerlayers=dict(
                     type='BaseTransformerLayer',
@@ -65,17 +65,19 @@
                     operation_order=('self_attn', 'norm', 'ffn', 'norm')),
                 init_cfg=None),
             positional_encoding=dict(
-                type='SinePositionalEncoding', num_feats=128, normalize=True),
+                type='mmdet.SinePositionalEncoding',
+                num_feats=128,
+                normalize=True),
             init_cfg=None),
         enforce_decoder_input_project=False,
         positional_encoding=dict(
             type='SinePositionalEncoding3D', num_feats=128, normalize=True),
         transformer_decoder=dict(
-            type='DetrTransformerDecoder',
+            type='mmdet.DetrTransformerDecoder',
             return_intermediate=True,
             num_layers=9,
             transformerlayers=dict(
-                type='DetrTransformerDecoderLayer',
+                type='mmdet.DetrTransformerDecoderLayer',
                 attn_cfgs=dict(
                     type='MultiheadAttention',
                     embed_dims=256,

diff --git a/mmtrack/apis/inference.py b/mmtrack/apis/inference.py
@@ -27,7 +27,7 @@ def init_model(config: Union[str, mmengine.Config],
     """Initialize a model from config file.
 
     Args:
-        config (str or :obj:`mmcv.Config`): Config file path or the config
+        config (str or :obj:`mmengine.Config`): Config file path or the config
             object.
         checkpoint (Optional[str], optional): Checkpoint path. Defaults to
             None.

diff --git a/mmtrack/datasets/transforms/processing.py b/mmtrack/datasets/transforms/processing.py
@@ -5,7 +5,7 @@
 
 import numpy as np
 from mmcv.transforms import BaseTransform
-from mmcv.utils import print_log
+from mmengine.logging import print_log
 
 from mmtrack.registry import TRANSFORMS
 

diff --git a/mmtrack/datasets/transforms/transforms.py b/mmtrack/datasets/transforms/transforms.py
@@ -6,7 +6,7 @@
 import numpy as np
 from mmcv.transforms import BaseTransform
 from mmcv.transforms.utils import cache_randomness
-from mmcv.utils import print_log
+from mmengine.logging import print_log
 
 from mmtrack.registry import TRANSFORMS
 from mmtrack.utils import crop_image

diff --git a/mmtrack/engine/hooks/visualization_hook.py b/mmtrack/engine/hooks/visualization_hook.py
@@ -7,6 +7,7 @@
 from mmengine.fileio import FileClient
 from mmengine.hooks import Hook
 from mmengine.runner import Runner
+from mmengine.utils import mkdir_or_exist
 from mmengine.visualization import Visualizer
 
 from mmtrack.registry import HOOKS
@@ -122,7 +123,7 @@ def after_test_iter(self, runner: Runner, batch_idx: int,
         if self.test_out_dir is not None:
             self.test_out_dir = osp.join(runner.work_dir, runner.timestamp,
                                          self.test_out_dir)
-            mmcv.mkdir_or_exist(self.test_out_dir)
+            mkdir_or_exist(self.test_out_dir)
 
         assert len(data_batch) == len(outputs) == 1, \
             'only batch_size=1 is supported while testing.'

diff --git a/mmtrack/evaluation/metrics/base_video_metrics.py b/mmtrack/evaluation/metrics/base_video_metrics.py
@@ -7,10 +7,10 @@
 from typing import Optional
 
 import torch
-from mmengine import mkdir_or_exist
 from mmengine.dist import (barrier, broadcast, broadcast_object_list,
                            get_dist_info, is_main_process)
 from mmengine.evaluator import BaseMetric
+from mmengine.utils import mkdir_or_exist
 
 
 class BaseVideoMetric(BaseMetric):

diff --git a/mmtrack/models/track_heads/mask2former_head.py b/mmtrack/models/track_heads/mask2former_head.py
@@ -5,16 +5,16 @@
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
-from mmcv.cnn import Conv2d, build_plugin_layer, caffe2_xavier_init
-from mmcv.cnn.bricks.transformer import build_transformer_layer_sequence
+from mmcv.cnn import Conv2d
 from mmcv.ops import point_sample
-from mmcv.runner import ModuleList
 from mmdet.models.dense_heads import AnchorFreeHead
 from mmdet.models.dense_heads import MaskFormerHead as MMDET_MaskFormerHead
 from mmdet.models.utils import get_uncertain_point_coords_with_randomness
 from mmdet.structures.mask import mask2bbox
 from mmdet.utils import ConfigType, OptConfigType, OptMultiConfig, reduce_mean
 from mmengine.data import InstanceData
+from mmengine.model import ModuleList
+from mmengine.model.utils import caffe2_xavier_init
 from torch import Tensor
 
 from mmtrack.registry import MODELS, TASK_UTILS
@@ -115,10 +115,8 @@ def __init__(self,
             in_channels=in_channels,
             feat_channels=feat_channels,
             out_channels=out_channels)
-        # TODO: change to `MODELS.build`
-        self.pixel_decoder = build_plugin_layer(pixel_decoder_)[1]
-        self.transformer_decoder = build_transformer_layer_sequence(
-            transformer_decoder)
+        self.pixel_decoder = MODELS.build(pixel_decoder_)
+        self.transformer_decoder = MODELS.build(transformer_decoder)
         self.decoder_embed_dims = self.transformer_decoder.embed_dims
 
         self.decoder_input_projs = ModuleList()

diff --git a/mmtrack/models/track_heads/stark_head.py b/mmtrack/models/track_heads/stark_head.py
@@ -187,11 +187,11 @@ class StarkTransformer(Transformer):
     <https://arxiv.org/pdf/2005.12872>`_ for details.
 
     Args:
-        encoder (`mmcv.ConfigDict` | Dict): Config of
+        encoder (`mmengine.ConfigDict` | Dict): Config of
             TransformerEncoder. Defaults to None.
-        decoder ((`mmcv.ConfigDict` | Dict)): Config of
+        decoder ((`mmengine.ConfigDict` | Dict)): Config of
             TransformerDecoder. Defaults to None
-        init_cfg (obj:`mmcv.ConfigDict`): The Config for initialization.
+        init_cfg (obj:`mmengine.ConfigDict`): The Config for initialization.
             Defaults to None.
     """
 
@@ -274,23 +274,23 @@ class StarkHead(BaseModule):
 
     Args:
         num_query (int): Number of query in transformer.
-        transformer (obj:`mmcv.ConfigDict`|dict): Config for transformer.
+        transformer (obj:`mmengine.ConfigDict`|dict): Config for transformer.
             Default: None.
-        positional_encoding (obj:`mmcv.ConfigDict`|dict):
+        positional_encoding (obj:`mmengine.ConfigDict`|dict):
             Config for position encoding.
-        bbox_head (obj:`mmcv.ConfigDict`|dict, optional): Config for bbox head.
-            Defaults to None.
-        cls_head (obj:`mmcv.ConfigDict`|dict, optional): Config for
+        bbox_head (obj:`mmengine.ConfigDict`|dict, optional): Config for bbox
+            head. Defaults to None.
+        cls_head (obj:`mmengine.ConfigDict`|dict, optional): Config for
             classification head. Defaults to None.
-        loss_cls (obj:`mmcv.ConfigDict`|dict): Config of the
+        loss_cls (obj:`mmengine.ConfigDict`|dict): Config of the
             classification loss. Default `CrossEntropyLoss`.
-        loss_bbox (obj:`mmcv.ConfigDict`|dict): Config of the bbox
+        loss_bbox (obj:`mmengine.ConfigDict`|dict): Config of the bbox
             regression loss. Default `L1Loss`.
-        loss_iou (obj:`mmcv.ConfigDict`|dict): Config of the bbox
+        loss_iou (obj:`mmengine.ConfigDict`|dict): Config of the bbox
             regression iou loss. Default `GIoULoss`.
-        tran_cfg (obj:`mmcv.ConfigDict`|dict): Training config of
+        tran_cfg (obj:`mmengine.ConfigDict`|dict): Training config of
             transformer head.
-        test_cfg (obj:`mmcv.ConfigDict`|dict): Testing config of
+        test_cfg (obj:`mmengine.ConfigDict`|dict): Testing config of
             transformer head.
         init_cfg (dict or list[dict], optional): Initialization config dict.
             Default: None

diff --git a/mmtrack/structures/reid_data_sample.py b/mmtrack/structures/reid_data_sample.py
@@ -2,7 +2,7 @@
 from numbers import Number
 from typing import Sequence, Union
 
-import mmcv
+import mmengine
 import numpy as np
 import torch
 from mmengine.data import BaseDataElement, LabelData
@@ -30,7 +30,7 @@ def format_label(value: Union[torch.Tensor, np.ndarray, Sequence, int],
 
     if isinstance(value, np.ndarray):
         value = torch.from_numpy(value)
-    elif isinstance(value, Sequence) and not mmcv.is_str(value):
+    elif isinstance(value, Sequence) and not mmengine.utils.is_str(value):
         value = torch.tensor(value)
     elif isinstance(value, int):
         value = torch.LongTensor([value])

diff --git a/mmtrack/testing/_utils.py b/mmtrack/testing/_utils.py
@@ -64,7 +64,7 @@ def _get_config_directory():
 
 def _get_config_module(fname):
     """Load a configuration as a python module."""
-    from mmcv import Config
+    from mmengine import Config
     config_dpath = _get_config_directory()
     config_fpath = join(config_dpath, fname)
     config_mod = Config.fromfile(config_fpath)

diff --git a/mmtrack/utils/collect_env.py b/mmtrack/utils/collect_env.py
@@ -1,6 +1,6 @@
 # Copyright (c) OpenMMLab. All rights reserved.
 from mmcv.utils import collect_env as collect_base_env
-from mmcv.utils import get_git_hash
+from mmengine.utils import get_git_hash
 
 import mmtrack
 

diff --git a/mmtrack/utils/mot_error_visualization.py b/mmtrack/utils/mot_error_visualization.py
@@ -8,7 +8,7 @@
 import numpy as np
 import seaborn as sns
 from matplotlib.patches import Rectangle
-from mmcv.utils import mkdir_or_exist
+from mmengine.utils import mkdir_or_exist
 
 
 def imshow_mot_errors(*args, backend: str = 'cv2', **kwargs):

diff --git a/tests/data/demo_cocovid_data/create_assets.py b/tests/data/demo_cocovid_data/create_assets.py
@@ -2,7 +2,8 @@
 from collections import defaultdict
 
 import mmcv
-from mmdet.core.bbox.demodata import random_boxes
+
+from mmtrack.testing import random_boxes
 
 
 def create_dummy_data():

diff --git a/tests/test_models/test_filter/test_filter_head.py b/tests/test_models/test_filter/test_filter_head.py
@@ -1,13 +1,12 @@
 # Copyright (c) OpenMMLab. All rights reserved.
-import pytest
 import torch
 
 from mmtrack.models import FilterInitializer
 
 
-@pytest.mark.skipif(
-    not torch.cuda.is_available, reason='test case under gpu environment')
 def test_filter_classifier_initializer():
+    if not torch.cuda.is_available():
+        return
     classifier_initializer = FilterInitializer(
         filter_size=4, feature_dim=8, feature_stride=16).to('cuda:0')
     feats = torch.randn(4, 8, 22, 22, device='cuda:0')

diff --git a/tests/test_models/test_mot/test_strong_sort.py b/tests/test_models/test_mot/test_strong_sort.py
@@ -24,6 +24,10 @@ def setUpClass(cls):
     ])
     def test_init(self, cfg_file):
         model = get_model_cfg(cfg_file)
+        model.reid.backbone.depth = 18
+        model.reid.head.fc_channels = 1
+        model.reid.head.out_channels = 1
+        model.reid.head.num_classes = 2
         model = MODELS.build(model)
         assert model.detector
         assert model.reid

diff --git a/tests/test_models/test_reid/test_base_reid.py b/tests/test_models/test_reid/test_base_reid.py
@@ -17,7 +17,7 @@ def setUpClass(cls) -> None:
         register_all_modules()
 
     @parameterized.expand([
-        'reid/reid_resnet50_8x32bs-6e_mot17train80_test-mot17val20.py',
+        'reid/reid_r50_8xb32-6e_mot17train80_test-mot17val20.py',
     ])
     def test_forward(self, cfg_file):
         model_cfg = get_model_cfg(cfg_file)

diff --git a/tests/test_models/test_roi_heads/test_bbox_heads/test_selsa_bbox_head.py b/tests/test_models/test_roi_heads/test_bbox_heads/test_selsa_bbox_head.py
@@ -23,6 +23,8 @@ def setUpClass(cls):
                 target_means=[0., 0., 0., 0.],
                 target_stds=[0.2, 0.2, 0.2, 0.2]),
             reg_class_agnostic=False,
+            reg_predictor_cfg=dict(type='mmdet.Linear'),
+            cls_predictor_cfg=dict(type='mmdet.Linear'),
             loss_cls=dict(
                 type='mmdet.CrossEntropyLoss',
                 use_sigmoid=False,

diff --git a/tests/test_models/test_sot/test_prdimp.py b/tests/test_models/test_sot/test_prdimp.py
@@ -1,7 +1,6 @@
 # Copyright (c) OpenMMLab. All rights reserved.
 from unittest import TestCase
 
-import pytest
 import torch
 from parameterized import parameterized
 
@@ -17,7 +16,7 @@ class TestPrDiMP(TestCase):
     def setUpClass(cls):
         register_all_modules(init_default_scope=True)
 
-    @parameterized.expand(['sot/prdimp/prdimp_resnet50_8x10bs-50e_got10k.py'])
+    @parameterized.expand(['sot/prdimp/prdimp_r50_8xb10-50e_got10k.py'])
     def test_init(self, cfg_file):
         model = get_model_cfg(cfg_file)
 
@@ -26,10 +25,11 @@ def test_init(self, cfg_file):
         assert model.classifier
         assert model.bbox_regressor
 
-    @pytest.mark.skipif(
-        not torch.cuda.is_available, reason='test case under gpu environment')
-    @parameterized.expand(['sot/prdimp/prdimp_resnet50_8x10bs-50e_got10k.py'])
+    @parameterized.expand(['sot/prdimp/prdimp_r50_8xb10-50e_got10k.py'])
     def test_stark_forward_predict_mode(self, cfg_file):
+        if not torch.cuda.is_available():
+            return
+
         _model = get_model_cfg(cfg_file)
         model = MODELS.build(_model)
         model = model.cuda()

diff --git a/tests/test_models/test_sot/test_siamrpn.py b/tests/test_models/test_sot/test_siamrpn.py
@@ -18,7 +18,7 @@ def setUpClass(cls):
         register_all_modules(init_default_scope=True)
 
     @parameterized.expand([
-        'sot/siamese_rpn/siamese-rpn_resnet50_8xb28-20e_imagenetvid-imagenetdet-coco_test-lasot.py',  # noqa: E501
+        'sot/siamese_rpn/siamese-rpn_r50_8xb28-20e_imagenetvid-imagenetdet-coco_test-lasot.py',  # noqa: E501
     ])
     def test_init(self, cfg_file):
         model = get_model_cfg(cfg_file)
@@ -30,7 +30,7 @@ def test_init(self, cfg_file):
 
     @parameterized.expand([
         (
-            'sot/siamese_rpn/siamese-rpn_resnet50_8xb28-20e_imagenetvid-imagenetdet-coco_test-lasot.py',  # noqa: E501
+            'sot/siamese_rpn/siamese-rpn_r50_8xb28-20e_imagenetvid-imagenetdet-coco_test-lasot.py',  # noqa: E501
             ('cpu', 'cuda')),
     ])
     def test_siamese_rpn_forward_loss_mode(self, cfg_file, devices):
@@ -61,7 +61,7 @@ def test_siamese_rpn_forward_loss_mode(self, cfg_file, devices):
 
     @parameterized.expand([
         (
-            'sot/siamese_rpn/siamese-rpn_resnet50_8xb28-20e_imagenetvid-imagenetdet-coco_test-lasot.py',  # noqa: E501
+            'sot/siamese_rpn/siamese-rpn_r50_8xb28-20e_imagenetvid-imagenetdet-coco_test-lasot.py',  # noqa: E501
             ('cpu', 'cuda')),
     ])
     def test_siamese_rpn_forward_predict_mode(self, cfg_file, devices):