add more results of OCRNet (#20)

* update the HRNet-OCR & add ResNet-101-OCR

* revise the script

* add the results of resnet-101+ocr

* add cascade ocr, aspp ocr

* add comparison table

* move comparison table

* support ocr+decoder

* revise the ocrnet_sep_aspp

* update the results of ocrnet

* update the results of ocrnet

* add sep-ocr-variants

* add bs2x exp of deeplabv3/v3+

* apply sep-conv in ocr module

* update the results

* update the results

* update the results of OCRNet

* update the results of OCRNet

* correct the results

* verify the release branch

* init the release branch

* add more results of ocrnet and ocrnetplus

* resolve the conflicts

* rename OCRNetPlus as OCRNet+

* fix the format

* fix the lint issues

* fix the lint issues

* fix the lint issues

* fix the lint isort issues

* fix the lint yapf issues

* fix the format issues

* remove the changes by the master branch

* remove the changes by the master branch

* remove the changes by the master branch

* remove the changes by the master branch

* remove the changes by the master branch

* add the logs folder to .gitignore

* recover .gitignore

* update readme

* update readme

* reset the cudnn_benchmark

* revise the README of OCRNet

* revise the name

* revise the reference of OCRNet

* revise the Figure of OCRNet+

* update the results of OCR/OCR+

* update the results of OCR/OCR+

* update the results of OCR/OCR+

* fix the format issue

* fix the format issue

* remove the ocr+

* update the results

* update the results

* fix the conflicts

* fix the lint issue

* fix the lint issue

* fix the lint issue

* fix the lint issue

* fix the inconsistency

* add urls to README

* clean the code

* remove the schedule configs

* clean the custom code

* clean up

* remove ocr.png

Co-authored-by: Jiarui XU <xvjiarui0826@gmail.com>

configs/_base_/models/ocrnet_r50-d8.py

@@ -0,0 +1,47 @@
+# model settings
+norm_cfg = dict(type='SyncBN', requires_grad=True)
+model = dict(
+    type='CascadeEncoderDecoder',
+    num_stages=2,
+    pretrained='open-mmlab://resnet50_v1c',
+    backbone=dict(
+        type='ResNetV1c',
+        depth=50,
+        num_stages=4,
+        out_indices=(0, 1, 2, 3),
+        dilations=(1, 1, 2, 4),
+        strides=(1, 2, 1, 1),
+        norm_cfg=norm_cfg,
+        norm_eval=False,
+        style='pytorch',
+        contract_dilation=True),
+    decode_head=[
+        dict(
+            type='FCNHead',
+            in_channels=1024,
+            in_index=2,
+            channels=256,
+            num_convs=1,
+            concat_input=False,
+            drop_out_ratio=0.1,
+            num_classes=19,
+            norm_cfg=norm_cfg,
+            align_corners=False,
+            loss_decode=dict(
+                type='CrossEntropyLoss', use_sigmoid=False, loss_weight=0.4)),
+        dict(
+            type='OCRHead',
+            in_channels=2048,
+            in_index=3,
+            channels=512,
+            ocr_channels=256,
+            drop_out_ratio=0.1,
+            num_classes=19,
+            norm_cfg=norm_cfg,
+            align_corners=False,
+            loss_decode=dict(
+                type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0))
+    ])
+# model training and testing settings
+train_cfg = dict()
+test_cfg = dict(mode='whole')

configs/ocrnet/README.md

@@ -1,18 +1,28 @@
 # Object-Contextual Representations for Semantic Segmentation
 
 ## Introduction
+
 ```
-@article{yuan2019ocr,
+@article{YuanW18,
+  title={Ocnet: Object context network for scene parsing},
+  author={Yuhui Yuan and Jingdong Wang},
+  booktitle={arXiv preprint arXiv:1809.00916},
+  year={2018}
+}
+
+@article{YuanCW20,
   title={Object-Contextual Representations for Semantic Segmentation},
-  author={Yuan Yuhui and Chen Xilin and Wang Jingdong},
+  author={Yuhui Yuan and Xilin Chen and Jingdong Wang},
-  journal={arXiv preprint arXiv:1909.11065},
+  booktitle={ECCV},
-  year={2019}
+  year={2020}
 }
 ```
 
 ## Results and models
 
 ### Cityscapes
+
+#### HRNet backbone
 | Method |      Backbone      | Crop Size | Lr schd | Mem (GB) | Inf time (fps) | mIoU  | mIoU(ms+flip) |                                                                                                                                                                                               download                                                                                                                                                                                               |
 |--------|--------------------|-----------|--------:|----------|----------------|------:|--------------:|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
 | OCRNet | HRNetV2p-W18-Small | 512x1024  |   40000 |      3.5 |          10.45 | 74.30 |         75.95 | [model](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/ocrnet/ocrnet_hr18s_512x1024_40k_cityscapes/ocrnet_hr18s_512x1024_40k_cityscapes_20200601_033304-fa2436c2.pth) | [log](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/ocrnet/ocrnet_hr18s_512x1024_40k_cityscapes/ocrnet_hr18s_512x1024_40k_cityscapes_20200601_033304.log.json)     |
@@ -25,6 +35,16 @@
 | OCRNet | HRNetV2p-W18       | 512x1024  |  160000 | -        | -              | 79.47 |         80.91 | [model](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/ocrnet/ocrnet_hr18_512x1024_160k_cityscapes/ocrnet_hr18_512x1024_160k_cityscapes_20200602_191001-b9172d0c.pth) | [log](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/ocrnet/ocrnet_hr18_512x1024_160k_cityscapes/ocrnet_hr18_512x1024_160k_cityscapes_20200602_191001.log.json)     |
 | OCRNet | HRNetV2p-W48       | 512x1024  |  160000 | -        | -              | 81.35 |         82.70 | [model](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/ocrnet/ocrnet_hr48_512x1024_160k_cityscapes/ocrnet_hr48_512x1024_160k_cityscapes_20200602_191037-dfbf1b0c.pth) | [log](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/ocrnet/ocrnet_hr48_512x1024_160k_cityscapes/ocrnet_hr48_512x1024_160k_cityscapes_20200602_191037.log.json)     |
 
+
+#### ResNet backbone
+
+| Method |      Backbone      | Crop Size | Batch Size | Lr schd | Mem (GB) | Inf time (fps) | mIoU  | mIoU(ms+flip) |                                                                                                                                                                                               download                                                                                                                                                                                               |
+|--------|--------------------|-----------|--------|----------|-----------|----------------|------|--------------:|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+| OCRNet  | R-101-D8 | 512x1024  | 8 |   40000 |  -   |   -  |   80.09  |  -  | [model](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/ocrnet/ocrnet_r101-d8_512x1024_40k_b8_cityscapes/ocrnet_r101-d8_512x1024_40k_b8_cityscapes-02ac0f13.pth) | [log](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/ocrnet/ocrnet_r101-d8_512x1024_40k_b8_cityscapes/ocrnet_r101-d8_512x1024_40k_b8_cityscapes_20200717_110721.log.json) |
+| OCRNet  | R-101-D8 | 512x1024  | 16 |   40000 |  8.8   |   3.02  |   80.30  |  -  | [model](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/ocrnet/ocrnet_r101-d8_512x1024_40k_b16_cityscapes/ocrnet_r101-d8_512x1024_40k_b16_cityscapes-db500f80.pth) | [log](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/ocrnet/ocrnet_r101-d8_512x1024_40k_b16_cityscapes/ocrnet_r101-d8_512x1024_40k_b16_cityscapes_20200723_193726.log.json) |
+| OCRNet  | R-101-D8 | 512x1024  | 16 |   80000 |  8.8   |   3.02  |   80.81  |  -  | [model](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/ocrnet/ocrnet_r101-d8_512x1024_80k_b16_cityscapes/ocrnet_r101-d8_512x1024_80k_b16_cityscapes-78688424.pth) | [log](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/ocrnet/ocrnet_r101-d8_512x1024_80k_b16_cityscapes/ocrnet_r101-d8_512x1024_80k_b16_cityscapes_20200723_192421.log.json) |
+
+
 ### ADE20K
 | Method |      Backbone      | Crop Size | Lr schd | Mem (GB) | Inf time (fps) | mIoU  | mIoU(ms+flip) |                                                                                                                                                                                     download                                                                                                                                                                                     |
 |--------|--------------------|-----------|--------:|----------|----------------|------:|--------------:|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|

configs/ocrnet/ocrnet_r101-d8_512x1024_40k_b16_cityscapes.py

@@ -0,0 +1,9 @@
+_base_ = [
+    '../_base_/models/ocrnet_r50-d8.py',
+    '../_base_/datasets/cityscapes.py',
+    '../_base_/default_runtime.py',
+    '../_base_/schedules/schedule_40k.py'
+]
+model = dict(pretrained='open-mmlab://resnet101_v1c', backbone=dict(depth=101))
+optimizer = dict(lr=0.02)
+lr_config = dict(min_lr=2e-4)

configs/ocrnet/ocrnet_r101-d8_512x1024_40k_b8_cityscapes.py

@@ -0,0 +1,7 @@
+_base_ = [
+    '../_base_/models/ocrnet_r50-d8.py',
+    '../_base_/datasets/cityscapes.py',
+    '../_base_/default_runtime.py',
+    '../_base_/schedules/schedule_40k.py'
+]
+model = dict(pretrained='open-mmlab://resnet101_v1c', backbone=dict(depth=101))

configs/ocrnet/ocrnet_r101-d8_512x1024_80k_b16_cityscapes.py

@@ -0,0 +1,9 @@
+_base_ = [
+    '../_base_/models/ocrnet_r50-d8.py',
+    '../_base_/datasets/cityscapes.py',
+    '../_base_/default_runtime.py',
+    '../_base_/schedules/schedule_80k.py'
+]
+model = dict(pretrained='open-mmlab://resnet101_v1c', backbone=dict(depth=101))
+optimizer = dict(lr=0.02)
+lr_config = dict(min_lr=2e-4)