(OLD) [Feature] Support LoveDA dataset

.github/workflows/build.yml

@@ -71,9 +71,17 @@ jobs:
         run: rm -rf .eggs && pip install -e .
       - name: Run unittests and generate coverage report
         run: |
+          pip install timm
           coverage run --branch --source mmseg -m pytest tests/
           coverage xml
           coverage report -m
+        if: ${{matrix.torch >= '1.5.0'}}
+      - name: Skip timm unittests and generate coverage report
+        run: |
+          coverage run --branch --source mmseg -m pytest tests/ --ignore tests/test_models/test_backbones/test_timm_backbone.py
+          coverage xml
+          coverage report -m
+        if: ${{matrix.torch < '1.5.0'}}
 
   build_cuda101:
     runs-on: ubuntu-18.04
@@ -142,9 +150,17 @@ jobs:
           TORCH_CUDA_ARCH_LIST=7.0 pip install .
       - name: Run unittests and generate coverage report
         run: |
+          python -m pip install timm
           coverage run --branch --source mmseg -m pytest tests/
           coverage xml
           coverage report -m
+        if: ${{matrix.torch >= '1.5.0'}}
+      - name: Skip timm unittests and generate coverage report
+        run: |
+          coverage run --branch --source mmseg -m pytest tests/ --ignore tests/test_models/test_backbones/test_timm_backbone.py
+          coverage xml
+          coverage report -m
+        if: ${{matrix.torch < '1.5.0'}}
       - name: Upload coverage to Codecov
         uses: codecov/codecov-action@v1.0.10
         with:
@@ -198,6 +214,7 @@ jobs:
           TORCH_CUDA_ARCH_LIST=7.0 pip install .
       - name: Run unittests and generate coverage report
         run: |
+          python -m pip install timm
           coverage run --branch --source mmseg -m pytest tests/
           coverage xml
           coverage report -m

README.md

@@ -49,7 +49,7 @@ This project is released under the [Apache 2.0 license](LICENSE).
 
 ## Changelog
 
-v0.18.0 was released in 10/07/2021.
+v0.19.0 was released in 11/02/2021.
 Please refer to [changelog.md](docs/changelog.md) for details and release history.
 
 ## Benchmark and model zoo

README_zh-CN.md

@@ -48,7 +48,7 @@ MMSegmentation 是一个基于 PyTorch 的语义分割开源工具箱。它是 O
 
 ## 更新日志
 
-最新的月度版本 v0.18.0 在 2021.10.07 发布。
+最新的月度版本 v0.19.0 在 2021.11.2 发布。
 如果想了解更多版本更新细节和历史信息，请阅读[更新日志](docs/changelog.md)。
 
 ## 基准测试和模型库

configs/_base_/datasets/loveda.py

@@ -0,0 +1,54 @@
+# dataset settings
+dataset_type = 'LoveDADataset'
+data_root = 'data/loveDA'
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+crop_size = (512, 512)
+train_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(type='LoadAnnotations', reduce_zero_label=True),
+    dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),
+    dict(type='RandomCrop', crop_size=crop_size, cat_max_ratio=0.75),
+    dict(type='RandomFlip', prob=0.5),
+    dict(type='PhotoMetricDistortion'),
+    dict(type='Normalize', **img_norm_cfg),
+    dict(type='Pad', size=crop_size, pad_val=0, seg_pad_val=255),
+    dict(type='DefaultFormatBundle'),
+    dict(type='Collect', keys=['img', 'gt_semantic_seg']),
+]
+test_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(
+        type='MultiScaleFlipAug',
+        img_scale=(1024, 1024),
+        # img_ratios=[0.5, 0.75, 1.0, 1.25, 1.5, 1.75],
+        flip=False,
+        transforms=[
+            dict(type='Resize', keep_ratio=True),
+            dict(type='RandomFlip'),
+            dict(type='Normalize', **img_norm_cfg),
+            dict(type='ImageToTensor', keys=['img']),
+            dict(type='Collect', keys=['img']),
+        ])
+]
+data = dict(
+    samples_per_gpu=4,
+    workers_per_gpu=4,
+    train=dict(
+        type=dataset_type,
+        data_root=data_root,
+        img_dir='img_dir/train',
+        ann_dir='ann_dir/train',
+        pipeline=train_pipeline),
+    val=dict(
+        type=dataset_type,
+        data_root=data_root,
+        img_dir='img_dir/val',
+        ann_dir='ann_dir/val',
+        pipeline=test_pipeline),
+    test=dict(
+        type=dataset_type,
+        data_root=data_root,
+        img_dir='img_dir/test',
+        ann_dir=None,
+        pipeline=test_pipeline))

configs/deeplabv3plus/deeplabv3plus_r101-d8_512x512_80k_loveda.py

@@ -0,0 +1,2 @@
+_base_ = './deeplabv3plus_r50-d8_512x512_80k_loveda.py'
+model = dict(pretrained='open-mmlab://resnet101_v1c', backbone=dict(depth=101))

configs/deeplabv3plus/deeplabv3plus_r18-d8_512x512_80k_loveda.py

@@ -0,0 +1,11 @@
+_base_ = './deeplabv3plus_r50-d8_512x512_80k_loveda.py'
+model = dict(
+    pretrained='open-mmlab://resnet18_v1c',
+    backbone=dict(depth=18),
+    decode_head=dict(
+        c1_in_channels=64,
+        c1_channels=12,
+        in_channels=512,
+        channels=128,
+    ),
+    auxiliary_head=dict(in_channels=256, channels=64))

configs/deeplabv3plus/deeplabv3plus_r50-d8_512x512_80k_loveda.py

@@ -0,0 +1,6 @@
+_base_ = [
+    '../_base_/models/deeplabv3plus_r50-d8.py', '../_base_/datasets/loveda.py',
+    '../_base_/default_runtime.py', '../_base_/schedules/schedule_80k.py'
+]
+model = dict(
+    decode_head=dict(num_classes=7), auxiliary_head=dict(num_classes=7))

configs/hrnet/fcn_hr18_512x512_80k_loveda.py

@@ -0,0 +1,4 @@
+_base_ = [
+    '../_base_/models/fcn_hr18.py', '../_base_/datasets/loveda.py',
+    '../_base_/default_runtime.py', '../_base_/schedules/schedule_80k.py'
+]

configs/hrnet/fcn_hr18s_512x512_80k_loveda.py

@@ -0,0 +1,10 @@
+_base_ = './fcn_hr18_512x512_80k_loveda.py'
+model = dict(
+    # pretrained='open-mmlab://msra/hrnetv2_w18_small',
+    pretrained='./pretrained/hrnetv2_w18_small-b5a04e21.pth',
+    backbone=dict(
+        extra=dict(
+            stage1=dict(num_blocks=(2, )),
+            stage2=dict(num_blocks=(2, 2)),
+            stage3=dict(num_modules=3, num_blocks=(2, 2, 2)),
+            stage4=dict(num_modules=2, num_blocks=(2, 2, 2, 2)))))

configs/hrnet/fcn_hr48_512x512_80k_loveda.py

@@ -0,0 +1,11 @@
+_base_ = './fcn_hr18_512x512_80k_loveda.py'
+model = dict(
+    # pretrained='open-mmlab://msra/hrnetv2_w48',
+    pretrained='./pretrained/hrnetv2_w48-d2186c55.pth',
+    backbone=dict(
+        extra=dict(
+            stage2=dict(num_channels=(48, 96)),
+            stage3=dict(num_channels=(48, 96, 192)),
+            stage4=dict(num_channels=(48, 96, 192, 384)))),
+    decode_head=dict(
+        in_channels=[48, 96, 192, 384], channels=sum([48, 96, 192, 384])))

configs/pspnet/pspnet_r101-d8_512x512_80k_loveda.py

@@ -0,0 +1,2 @@
+_base_ = './pspnet_r50-d8_512x512_80k_loveda.py'
+model = dict(pretrained='open-mmlab://resnet101_v1c', backbone=dict(depth=101))

configs/pspnet/pspnet_r18-d8_512x512_80k_loveda.py

@@ -0,0 +1,9 @@
+_base_ = './pspnet_r50-d8_512x512_80k_loveda.py'
+model = dict(
+    pretrained='open-mmlab://resnet18_v1c',
+    backbone=dict(depth=18),
+    decode_head=dict(
+        in_channels=512,
+        channels=128,
+    ),
+    auxiliary_head=dict(in_channels=256, channels=64))

configs/pspnet/pspnet_r50-d8_512x512_80k_loveda.py

@@ -0,0 +1,6 @@
+_base_ = [
+    '../_base_/models/pspnet_r50-d8.py', '../_base_/datasets/loveda.py',
+    '../_base_/default_runtime.py', '../_base_/schedules/schedule_80k.py'
+]
+model = dict(
+    decode_head=dict(num_classes=7), auxiliary_head=dict(num_classes=7))

docs/changelog.md

@@ -1,5 +1,45 @@
 ## Changelog
 
+### V0.19 (11/02/2021)
+
+**Highlights**
+
+- Support TIMMBackbone wrapper ([#998](https://github.com/open-mmlab/mmsegmentation/pull/998))
+- Support custom hook ([#428](https://github.com/open-mmlab/mmsegmentation/pull/428))
+- Add codespell pre-commit hook ([#920](https://github.com/open-mmlab/mmsegmentation/pull/920))
+- Add FastFCN benchmark on ADE20K ([#972](https://github.com/open-mmlab/mmsegmentation/pull/972))
+
+**New Features**
+
+- Support TIMMBackbone wrapper ([#998](https://github.com/open-mmlab/mmsegmentation/pull/998))
+- Support custom hook ([#428](https://github.com/open-mmlab/mmsegmentation/pull/428))
+- Add FastFCN benchmark on ADE20K ([#972](https://github.com/open-mmlab/mmsegmentation/pull/972))
+- Add codespell pre-commit hook and fix typos ([#920](https://github.com/open-mmlab/mmsegmentation/pull/920))
+
+**Improvements**
+
+- Make inputs & channels smaller in unittests ([#1004](https://github.com/open-mmlab/mmsegmentation/pull/1004))
+- Change `self.loss_decode` back to `dict` in Single Loss situation ([#1002](https://github.com/open-mmlab/mmsegmentation/pull/1002))
+
+**Bug Fixes**
+
+- Fix typo in usage example ([#1003](https://github.com/open-mmlab/mmsegmentation/pull/1003))
+- Add contiguous after permutation in ViT ([#992](https://github.com/open-mmlab/mmsegmentation/pull/992))
+- Fix the invalid link ([#985](https://github.com/open-mmlab/mmsegmentation/pull/985))
+- Fix bug in CI with python 3.9 ([#994](https://github.com/open-mmlab/mmsegmentation/pull/994))
+- Fix bug when loading class name form file in custom dataset ([#923](https://github.com/open-mmlab/mmsegmentation/pull/923))
+
+**Contributors**
+
+- @ShoupingShan made their first contribution in https://github.com/open-mmlab/mmsegmentation/pull/923
+- @RockeyCoss made their first contribution in https://github.com/open-mmlab/mmsegmentation/pull/954
+- @HarborYuan made their first contribution in https://github.com/open-mmlab/mmsegmentation/pull/992
+- @lkm2835 made their first contribution in https://github.com/open-mmlab/mmsegmentation/pull/1003
+- @gszh made their first contribution in https://github.com/open-mmlab/mmsegmentation/pull/428
+- @VVsssssk
+- @MengzhangLI
+- @Junjun2016
+
 ### V0.18 (10/07/2021)
 
 **Highlights**

docs/dataset_prepare.md

@@ -253,3 +253,33 @@ Since we only support test models on this dataset, you may only download [the va
 ### Nighttime Driving
 
 Since we only support test models on this dataset, you may only download [the test set](http://data.vision.ee.ethz.ch/daid/NighttimeDriving/NighttimeDrivingTest.zip).
+
+### LoveDA
+
+The data could be downloaded [here](https://drive.google.com/drive/folders/1ibYV0qwn4yuuh068Rnc-w4tPi0U0c-ti?usp=sharing).
+
+For LoveDA dataset, please run the following command to download and re-organize the dataset.
+
+```shell
+# download
+mkdir loveda && cd loveda
+wget https://drive.google.com/drive/folders/1ibYV0qwn4yuuh068Rnc-w4tPi0U0c-ti?usp=sharing
+
+# unzip
+unzip '*.zip'
+
+# Convert into segmentation splits
+mkdir -p img_dir/train img_dir/val img_dir/test ann_dir/train ann_dir/val
+mv Train/Rural/images_png/* img_dir/train
+mv Train/Urban/images_png/* img_dir/train
+mv Val/Rural/images_png/* img_dir/val
+mv Val/Uban/images_png/* img_dir/val
+mv Test/Rural/images_png/* img_dir/val
+mv Test/Uban/images_png/* img_dir/val
+mv Train/Rural/masks_png/* ann_dir/train
+mv Train/Uban/masks_png/* ann_dir/train
+mv Val/Rural/masks_png/* ann_dir/val
+mv Val/Uban/masks_png/* ann_dir/val
+```
+
+More details about LoveDA can be found [here](https://github.com/Junjue-Wang/LoveDA).

docs/get_started.md

@@ -12,6 +12,7 @@ The compatible MMSegmentation and MMCV versions are as below. Please install the
 | MMSegmentation version |    MMCV version     |
 |:-------------------:|:-------------------:|
 | master              | mmcv-full>=1.3.13, <1.4.0 |
+| 0.19.0              | mmcv-full>=1.3.13, <1.4.0 |
 | 0.18.0              | mmcv-full>=1.3.13, <1.4.0 |
 | 0.17.0              | mmcv-full>=1.3.7, <1.4.0 |
 | 0.16.0              | mmcv-full>=1.3.7, <1.4.0 |

docs_zh-CN/dataset_prepare.md

@@ -195,3 +195,37 @@ python tools/convert_datasets/stare.py /path/to/stare-images.tar /path/to/labels
 ### Nighttime Driving
 
 因为我们只支持在此数据集上测试模型，所以您只需下载[测试集](http://data.vision.ee.ethz.ch/daid/NighttimeDriving/NighttimeDrivingTest.zip)。
+
+### LoveDA
+
+下载[LoveDA数据集](https://drive.google.com/drive/folders/1ibYV0qwn4yuuh068Rnc-w4tPi0U0c-ti?usp=sharing)。
+
+对于 LoveDA 数据集，请运行以下命令下载并组织数据集
+
+```shell
+
+# download
+
+mkdir loveda && cd loveda
+wget https://drive.google.com/drive/folders/1ibYV0qwn4yuuh068Rnc-w4tPi0U0c-ti?usp=sharing
+
+# unzip
+
+unzip '*.zip'
+
+# Convert into segmentation splits
+
+mkdir -p img_dir/train img_dir/val img_dir/test ann_dir/train ann_dir/val
+mv Train/Rural/images_png/* img_dir/train
+mv Train/Urban/images_png/* img_dir/train
+mv Val/Rural/images_png/* img_dir/val
+mv Val/Urban/images_png/* img_dir/val
+mv Test/Rural/images_png/* img_dir/val
+mv Test/Urban/images_png/* img_dir/val
+mv Train/Rural/masks_png/* ann_dir/train
+mv Train/Urban/masks_png/* ann_dir/train
+mv Val/Rural/masks_png/* ann_dir/val
+mv Val/Urban/masks_png/* ann_dir/val
+```
+
+关于 LoveDA 的更多细节可以在[这里](https://github.com/Junjue-Wang/LoveDA)找到。

docs_zh-CN/get_started.md

@@ -12,6 +12,7 @@
 | MMSegmentation 版本 |    MMCV 版本     |
 |:-------------------:|:-------------------:|
 | master              | mmcv-full>=1.3.13, <1.4.0 |
+| 0.19.0              | mmcv-full>=1.3.13, <1.4.0 |
 | 0.18.0              | mmcv-full>=1.3.13, <1.4.0 |
 | 0.17.0              | mmcv-full>=1.3.7, <1.4.0 |
 | 0.16.0              | mmcv-full>=1.3.7, <1.4.0 |

mmseg/core/seg/sampler/ohem_pixel_sampler.py

@@ -1,5 +1,6 @@
 # Copyright (c) OpenMMLab. All rights reserved.
 import torch
+import torch.nn as nn
 import torch.nn.functional as F
 
 from ..builder import PIXEL_SAMPLERS
@@ -62,14 +63,19 @@ def sample(self, seg_logit, seg_label):
                 threshold = max(min_threshold, self.thresh)
                 valid_seg_weight[seg_prob[valid_mask] < threshold] = 1.
             else:
+                if not isinstance(self.context.loss_decode, nn.ModuleList):
+                    losses_decode = [self.context.loss_decode]
+                else:
+                    losses_decode = self.context.loss_decode
                 losses = 0.0
-                for loss_module in self.context.loss_decode:
+                for loss_module in losses_decode:
                     losses += loss_module(
                         seg_logit,
                         seg_label,
                         weight=None,
                         ignore_index=self.context.ignore_index,
                         reduction_override='none')
+
                 # faster than topk according to https://github.com/pytorch/pytorch/issues/22812  # noqa
                 _, sort_indices = losses[valid_mask].sort(descending=True)
                 valid_seg_weight[sort_indices[:batch_kept]] = 1.

mmseg/datasets/__init__.py

@@ -9,16 +9,31 @@
 from .dataset_wrappers import ConcatDataset, RepeatDataset
 from .drive import DRIVEDataset
 from .hrf import HRFDataset
+from .loveda import LoveDADataset
 from .night_driving import NightDrivingDataset
 from .pascal_context import PascalContextDataset, PascalContextDataset59
 from .stare import STAREDataset
 from .voc import PascalVOCDataset
 
 __all__ = [
-    'CustomDataset', 'build_dataloader', 'ConcatDataset', 'RepeatDataset',
-    'DATASETS', 'build_dataset', 'PIPELINES', 'CityscapesDataset',
-    'PascalVOCDataset', 'ADE20KDataset', 'PascalContextDataset',
-    'PascalContextDataset59', 'ChaseDB1Dataset', 'DRIVEDataset', 'HRFDataset',
-    'STAREDataset', 'DarkZurichDataset', 'NightDrivingDataset',
-    'COCOStuffDataset'
+    'CustomDataset',
+    'build_dataloader',
+    'ConcatDataset',
+    'RepeatDataset',
+    'DATASETS',
+    'build_dataset',
+    'PIPELINES',
+    'CityscapesDataset',
+    'PascalVOCDataset',
+    'ADE20KDataset',
+    'PascalContextDataset',
+    'PascalContextDataset59',
+    'ChaseDB1Dataset',
+    'DRIVEDataset',
+    'HRFDataset',
+    'STAREDataset',
+    'DarkZurichDataset',
+    'NightDrivingDataset',
+    'COCOStuffDataset',
+    'LoveDADataset',
 ]