[Feature] Support iSAID aerial dataset.

README.md

@@ -138,6 +138,7 @@ Supported datasets:
 - [x] [LoveDA](https://github.com/open-mmlab/mmsegmentation/blob/master/docs/en/dataset_prepare.md#loveda)
 - [x] [Potsdam](https://github.com/open-mmlab/mmsegmentation/blob/master/docs/en/dataset_prepare.md#isprs-potsdam)
 - [x] [Vaihingen](https://github.com/open-mmlab/mmsegmentation/blob/master/docs/en/dataset_prepare.md#isprs-vaihingen)
+- [x] [iSAID](https://github.com/open-mmlab/mmsegmentation/blob/master/docs/en/dataset_prepare.md#isaid)
 
 ## Installation
 

README_zh-CN.md

@@ -137,6 +137,7 @@ MMSegmentation 是一个基于 PyTorch 的语义分割开源工具箱。它是 O
 - [x] [LoveDA](https://github.com/open-mmlab/mmsegmentation/blob/master/docs/zh_cn/dataset_prepare.md#loveda)
 - [x] [Potsdam](https://github.com/open-mmlab/mmsegmentation/blob/master/docs/zh_cn/dataset_prepare.md#isprs-potsdam)
 - [x] [Vaihingen](https://github.com/open-mmlab/mmsegmentation/blob/master/docs/zh_cn/dataset_prepare.md#isprs-vaihingen)
+- [x] [iSAID](https://github.com/open-mmlab/mmsegmentation/blob/master/docs/zh_cn/dataset_prepare.md#isaid)
 
 ## 安装
 

configs/_base_/datasets/isaid.py

@@ -0,0 +1,62 @@
+# dataset settings
+dataset_type = 'iSAIDDataset'
+data_root = 'data/iSAID'
+
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+"""
+This crop_size setting is followed by the implementation of
+`PointFlow: Flowing Semantics Through Points for Aerial Image
+Segmentation <https://arxiv.org/pdf/2103.06564.pdf>`_.
+"""
+
+crop_size = (896, 896)
+
+train_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(type='LoadAnnotations'),
+    dict(type='Resize', img_scale=(896, 896), 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=(896, 896),
+        # 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/val',
+        ann_dir='ann_dir/val',
+        pipeline=test_pipeline))

configs/deeplabv3plus/README.md

@@ -114,8 +114,16 @@ Spatial pyramid pooling module or encode-decoder structure are used in deep neur
 | DeepLabV3+ | R-50-D8  | 512x512   |   80000 | 7.36      | 26.91             | 73.97 |         75.05 | [config](https://github.com/open-mmlab/mmsegmentation/blob/master/configs/deeplabv3plus/deeplabv3plus_r50-d8_4x4_512x512_80k_vaihingen.py)  | [model](https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3plus/deeplabv3plus_r50-d8_4x4_512x512_80k_vaihingen/deeplabv3plus_r50-d8_4x4_512x512_80k_vaihingen_20211231_230816-5040938d.pth) | [log](https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3plus/deeplabv3plus_r50-d8_4x4_512x512_80k_vaihingen/deeplabv3plus_r50-d8_4x4_512x512_80k_vaihingen_20211231_230816.log.json)     |
 | DeepLabV3+ | R-101-D8 | 512x512   |   80000 | 10.83       | 18.59          | 73.06 |         74.14 | [config](https://github.com/open-mmlab/mmsegmentation/blob/master/configs/deeplabv3plus/deeplabv3plus_r101-d8_4x4_512x512_80k_vaihingen.py) | [model](https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3plus/deeplabv3plus_r101-d8_4x4_512x512_80k_vaihingen/deeplabv3plus_r101-d8_4x4_512x512_80k_vaihingen_20211231_230816-8a095afa.pth) | [log](https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3plus/deeplabv3plus_r101-d8_4x4_512x512_80k_vaihingen/deeplabv3plus_r101-d8_4x4_512x512_80k_vaihingen_20211231_230816.log.json) |
 
+### iSAID
+
+| Method     | Backbone | Crop Size | Lr schd | Mem (GB) | Inf time (fps) |  mIoU | mIoU(ms+flip) | config                                                                                                                                 | download                                                                                                                                                                                                                                                                                                                                                                                   |
+| ---------- | -------- | --------- | ------: | -------- | -------------- | ----: | ------------: | -------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
+| DeepLabV3+ | R-18-D8  | 896x896   |   80000 | 6.19      | 24.81             | 61.35 |         62.61 | [config](https://github.com/open-mmlab/mmsegmentation/blob/master/configs/deeplabv3plus/deeplabv3plus_r18-d8_4x4_896x896_80k_isaid.py)  | [model](https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3plus/deeplabv3plus_r18-d8_4x4_896x896_80k_isaid/deeplabv3plus_r18-d8_4x4_896x896_80k_isaid_20220110_180526-7059991d.pth) | [log](https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3plus/deeplabv3plus_r18-d8_4x4_896x896_80k_isaid/deeplabv3plus_r18-d8_4x4_896x896_80k_isaid_20220110_180526.log.json)     |
+| DeepLabV3+ | R-50-D8  | 896x896   |   80000 | 21.45      | 8.42             | 67.06 |         68.02 | [config](https://github.com/open-mmlab/mmsegmentation/blob/master/configs/deeplabv3plus/deeplabv3plus_r50-d8_4x4_896x896_80k_isaid.py)  | [model](https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3plus/deeplabv3plus_r50-d8_4x4_896x896_80k_isaid/deeplabv3plus_r50-d8_4x4_896x896_80k_isaid_20220110_180526-598be439.pth) | [log](https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3plus/deeplabv3plus_r50-d8_4x4_896x896_80k_isaid/deeplabv3plus_r50-d8_4x4_896x896_80k_isaid_20220110_180526.log.json)     |
+
 Note:
 
 - `D-8`/`D-16` here corresponding to the output stride 8/16 setting for DeepLab series.
 - `MG-124` stands for multi-grid dilation in the last stage of ResNet.
 - `FP16` means Mixed Precision (FP16) is adopted in training.
+- `896x896` is the Crop Size of iSAID dataset, which is followed by the implementation of [PointFlow: Flowing Semantics Through Points for Aerial Image Segmentation](https://arxiv.org/pdf/2103.06564.pdf)

configs/deeplabv3plus/deeplabv3plus.yml

@@ -10,6 +10,7 @@ Collections:
     - LoveDA
     - Potsdam
     - Vaihingen
+    - iSAID
   Paper:
     URL: https://arxiv.org/abs/1802.02611
     Title: Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation
@@ -803,3 +804,47 @@ Models:
       mIoU(ms+flip): 74.14
   Config: configs/deeplabv3plus/deeplabv3plus_r101-d8_4x4_512x512_80k_vaihingen.py
   Weights: https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3plus/deeplabv3plus_r101-d8_4x4_512x512_80k_vaihingen/deeplabv3plus_r101-d8_4x4_512x512_80k_vaihingen_20211231_230816-8a095afa.pth
+- Name: deeplabv3plus_r18-d8_4x4_896x896_80k_isaid
+  In Collection: deeplabv3plus
+  Metadata:
+    backbone: R-18-D8
+    crop size: (896,896)
+    lr schd: 80000
+    inference time (ms/im):
+    - value: 40.31
+      hardware: V100
+      backend: PyTorch
+      batch size: 1
+      mode: FP32
+      resolution: (896,896)
+    Training Memory (GB): 6.19
+  Results:
+  - Task: Semantic Segmentation
+    Dataset: iSAID
+    Metrics:
+      mIoU: 61.35
+      mIoU(ms+flip): 62.61
+  Config: configs/deeplabv3plus/deeplabv3plus_r18-d8_4x4_896x896_80k_isaid.py
+  Weights: https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3plus/deeplabv3plus_r18-d8_4x4_896x896_80k_isaid/deeplabv3plus_r18-d8_4x4_896x896_80k_isaid_20220110_180526-7059991d.pth
+- Name: deeplabv3plus_r50-d8_4x4_896x896_80k_isaid
+  In Collection: deeplabv3plus
+  Metadata:
+    backbone: R-50-D8
+    crop size: (896,896)
+    lr schd: 80000
+    inference time (ms/im):
+    - value: 118.76
+      hardware: V100
+      backend: PyTorch
+      batch size: 1
+      mode: FP32
+      resolution: (896,896)
+    Training Memory (GB): 21.45
+  Results:
+  - Task: Semantic Segmentation
+    Dataset: iSAID
+    Metrics:
+      mIoU: 67.06
+      mIoU(ms+flip): 68.02
+  Config: configs/deeplabv3plus/deeplabv3plus_r50-d8_4x4_896x896_80k_isaid.py
+  Weights: https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3plus/deeplabv3plus_r50-d8_4x4_896x896_80k_isaid/deeplabv3plus_r50-d8_4x4_896x896_80k_isaid_20220110_180526-598be439.pth

configs/deeplabv3plus/deeplabv3plus_r18-d8_4x4_896x896_80k_isaid.py

@@ -0,0 +1,11 @@
+_base_ = './deeplabv3plus_r50-d8_4x4_896x896_80k_isaid.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_4x4_896x896_80k_isaid.py

@@ -0,0 +1,6 @@
+_base_ = [
+    '../_base_/models/deeplabv3plus_r50-d8.py', '../_base_/datasets/isaid.py',
+    '../_base_/default_runtime.py', '../_base_/schedules/schedule_80k.py'
+]
+model = dict(
+    decode_head=dict(num_classes=16), auxiliary_head=dict(num_classes=16))

configs/hrnet/README.md

@@ -107,3 +107,15 @@ High-resolution representations are essential for position-sensitive vision prob
 | FCN | HRNetV2p-W18-Small  | 512x512   |   80000 | 1.58      | 38.11             | 71.81 |         73.1 | [config](https://github.com/open-mmlab/mmsegmentation/blob/master/configs/hrnet/fcn_hr18s_4x4_512x512_80k_vaihingen.py)  | [model](https://download.openmmlab.com/mmsegmentation/v0.5/hrnet/fcn_hr18s_4x4_512x512_80k_vaihingen/fcn_hr18s_4x4_512x512_80k_vaihingen_20211231_230909-b23aae02.pth) | [log](https://download.openmmlab.com/mmsegmentation/v0.5/hrnet/fcn_hr18s_4x4_512x512_80k_vaihingen/fcn_hr18s_4x4_512x512_80k_vaihingen_20211231_230909.log.json)     |
 | FCN | HRNetV2p-W18  | 512x512   |   80000 | 2.76      | 19.55             | 72.57 |         74.09 | [config](https://github.com/open-mmlab/mmsegmentation/blob/master/configs/hrnet/fcn_hr18_4x4_512x512_80k_vaihingen.py)  | [model](https://download.openmmlab.com/mmsegmentation/v0.5/hrnet/fcn_hr18_4x4_512x512_80k_vaihingen/fcn_hr18_4x4_512x512_80k_vaihingen_20211231_231216-2ec3ae8a.pth) | [log](https://download.openmmlab.com/mmsegmentation/v0.5/hrnet/fcn_hr18_4x4_512x512_80k_vaihingen/fcn_hr18_4x4_512x512_80k_vaihingen_20211231_231216.log.json)     |
 | FCN | HRNetV2p-W48 | 512x512   |   80000 | 6.20       | 17.25          | 72.50 |         73.52 | [config](https://github.com/open-mmlab/mmsegmentation/blob/master/configs/hrnet/fcn_hr48_4x4_512x512_80k_vaihingen.py) | [model](https://download.openmmlab.com/mmsegmentation/v0.5/hrnet/fcn_hr48_4x4_512x512_80k_vaihingen/fcn_hr48_4x4_512x512_80k_vaihingen_20211231_231244-7133cb22.pth) | [log](https://download.openmmlab.com/mmsegmentation/v0.5/hrnet/fcn_hr48_4x4_512x512_80k_vaihingen/fcn_hr48_4x4_512x512_80k_vaihingen_20211231_231244.log.json) |
+
+### iSAID
+
+| Method     | Backbone | Crop Size | Lr schd | Mem (GB) | Inf time (fps) |  mIoU | mIoU(ms+flip) | config                                                                                                                                 | download                                                                                                                                                                                                                                                                                                                                                                                   |
+| ---------- | -------- | --------- | ------: | -------- | -------------- | ----: | ------------: | -------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
+| FCN | HRNetV2p-W18-Small  | 896x896   |   80000 | 4.95      | 13.84            | 62.30 |         62.97 | [config](https://github.com/open-mmlab/mmsegmentation/blob/master/configs/hrnet/fcn_hr18s_4x4_896x896_80k_isaid.py)  | [model](https://download.openmmlab.com/mmsegmentation/v0.5/hrnet/fcn_hr18s_4x4_896x896_80k_isaid/fcn_hr18s_4x4_896x896_80k_isaid_20220118_001603-3cc0769b.pth) | [log](https://download.openmmlab.com/mmsegmentation/v0.5/hrnet/fcn_hr18s_4x4_896x896_80k_isaid/fcn_hr18s_4x4_896x896_80k_isaid_20220118_001603.log.json)     |
+| FCN | HRNetV2p-W18  | 896x896   |   80000 | 8.30      | 7.71             | 65.06 |         65.60 | [config](https://github.com/open-mmlab/mmsegmentation/blob/master/configs/hrnet/fcn_hr18_4x4_896x896_80k_isaid.py)  | [model](https://download.openmmlab.com/mmsegmentation/v0.5/hrnet/fcn_hr18_4x4_896x896_80k_isaid/fcn_hr18_4x4_896x896_80k_isaid_20220110_182230-49bf752e.pth) | [log](https://download.openmmlab.com/mmsegmentation/v0.5/hrnet/fcn_hr18_4x4_896x896_80k_isaid/fcn_hr18_4x4_896x896_80k_isaid_20220110_182230.log.json)     |
+| FCN | HRNetV2p-W48 | 896x896   |   80000 | 16.89       | 7.34          | 67.80 |         68.53 | [config](https://github.com/open-mmlab/mmsegmentation/blob/master/configs/hrnet/fcn_hr48_4x4_896x896_80k_isaid.py) | [model](https://download.openmmlab.com/mmsegmentation/v0.5/hrnet/fcn_hr48_4x4_896x896_80k_isaid/fcn_hr48_4x4_896x896_80k_isaid_20220114_174643-547fc420.pth) | [log](https://download.openmmlab.com/mmsegmentation/v0.5/hrnet/fcn_hr48_4x4_896x896_80k_isaid/fcn_hr48_4x4_896x896_80k_isaid_20220114_174643.log.json) |
+
+Note:
+
+- `896x896` is the Crop Size of iSAID dataset, which is followed by the implementation of [PointFlow: Flowing Semantics Through Points for Aerial Image Segmentation](https://arxiv.org/pdf/2103.06564.pdf)

configs/hrnet/fcn_hr18_4x4_896x896_80k_isaid.py

@@ -0,0 +1,5 @@
+_base_ = [
+    '../_base_/models/fcn_hr18.py', '../_base_/datasets/isaid.py',
+    '../_base_/default_runtime.py', '../_base_/schedules/schedule_80k.py'
+]
+model = dict(decode_head=dict(num_classes=16))

configs/hrnet/fcn_hr18s_4x4_896x896_80k_isaid.py

@@ -0,0 +1,9 @@
+_base_ = './fcn_hr18_4x4_896x896_80k_isaid.py'
+model = dict(
+    pretrained='open-mmlab://msra/hrnetv2_w18_small',
+    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_4x4_896x896_80k_isaid.py

@@ -0,0 +1,10 @@
+_base_ = './fcn_hr18_4x4_896x896_80k_isaid.py'
+model = dict(
+    pretrained='open-mmlab://msra/hrnetv2_w48',
+    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/hrnet/hrnet.yml

@@ -10,6 +10,7 @@ Collections:
     - LoveDA
     - Potsdam
     - Vaihingen
+    - iSAID
   Paper:
     URL: https://arxiv.org/abs/1908.07919
     Title: Deep High-Resolution Representation Learning for Human Pose Estimation
@@ -648,3 +649,69 @@ Models:
       mIoU(ms+flip): 73.52
   Config: configs/hrnet/fcn_hr48_4x4_512x512_80k_vaihingen.py
   Weights: https://download.openmmlab.com/mmsegmentation/v0.5/hrnet/fcn_hr48_4x4_512x512_80k_vaihingen/fcn_hr48_4x4_512x512_80k_vaihingen_20211231_231244-7133cb22.pth
+- Name: fcn_hr18s_4x4_896x896_80k_isaid
+  In Collection: hrnet
+  Metadata:
+    backbone: HRNetV2p-W18-Small
+    crop size: (896,896)
+    lr schd: 80000
+    inference time (ms/im):
+    - value: 72.25
+      hardware: V100
+      backend: PyTorch
+      batch size: 1
+      mode: FP32
+      resolution: (896,896)
+    Training Memory (GB): 4.95
+  Results:
+  - Task: Semantic Segmentation
+    Dataset: iSAID
+    Metrics:
+      mIoU: 62.3
+      mIoU(ms+flip): 62.97
+  Config: configs/hrnet/fcn_hr18s_4x4_896x896_80k_isaid.py
+  Weights: https://download.openmmlab.com/mmsegmentation/v0.5/hrnet/fcn_hr18s_4x4_896x896_80k_isaid/fcn_hr18s_4x4_896x896_80k_isaid_20220118_001603-3cc0769b.pth
+- Name: fcn_hr18_4x4_896x896_80k_isaid
+  In Collection: hrnet
+  Metadata:
+    backbone: HRNetV2p-W18
+    crop size: (896,896)
+    lr schd: 80000
+    inference time (ms/im):
+    - value: 129.7
+      hardware: V100
+      backend: PyTorch
+      batch size: 1
+      mode: FP32
+      resolution: (896,896)
+    Training Memory (GB): 8.3
+  Results:
+  - Task: Semantic Segmentation
+    Dataset: iSAID
+    Metrics:
+      mIoU: 65.06
+      mIoU(ms+flip): 65.6
+  Config: configs/hrnet/fcn_hr18_4x4_896x896_80k_isaid.py
+  Weights: https://download.openmmlab.com/mmsegmentation/v0.5/hrnet/fcn_hr18_4x4_896x896_80k_isaid/fcn_hr18_4x4_896x896_80k_isaid_20220110_182230-49bf752e.pth
+- Name: fcn_hr48_4x4_896x896_80k_isaid
+  In Collection: hrnet
+  Metadata:
+    backbone: HRNetV2p-W48
+    crop size: (896,896)
+    lr schd: 80000
+    inference time (ms/im):
+    - value: 136.24
+      hardware: V100
+      backend: PyTorch
+      batch size: 1
+      mode: FP32
+      resolution: (896,896)
+    Training Memory (GB): 16.89
+  Results:
+  - Task: Semantic Segmentation
+    Dataset: iSAID
+    Metrics:
+      mIoU: 67.8
+      mIoU(ms+flip): 68.53
+  Config: configs/hrnet/fcn_hr48_4x4_896x896_80k_isaid.py
+  Weights: https://download.openmmlab.com/mmsegmentation/v0.5/hrnet/fcn_hr48_4x4_896x896_80k_isaid/fcn_hr48_4x4_896x896_80k_isaid_20220114_174643-547fc420.pth