CodeCamp #150 [Feature] Add ISNet (#2400)

## Motivation

Support ISNet.
paper link: [ISNet: Integrate Image-Level and Semantic-Level Context for
Semantic
Segmentation](https://openaccess.thecvf.com/content/ICCV2021/papers/Jin_ISNet_Integrate_Image-Level_and_Semantic-Level_Context_for_Semantic_Segmentation_ICCV_2021_paper.pdf)

## Modification

Add ISNet decoder head.
Add ISNet config.

projects/isnet/README.md

@@ -0,0 +1,57 @@
+# ISNet
+
+[ISNet: Integrate Image-Level and Semantic-Level Context for Semantic Segmentation](https://arxiv.org/pdf/2108.12382.pdf)
+
+## Description
+
+This is an implementation of [ISNet](https://arxiv.org/pdf/2108.12382.pdf).
+[Official Repo](https://github.com/SegmentationBLWX/sssegmentation)
+
+## Usage
+
+### Prerequisites
+
+- Python 3.7
+- PyTorch 1.6 or higher
+- [MIM](https://github.com/open-mmlab/mim) v0.33 or higher
+- [MMSegmentation](https://github.com/open-mmlab/mmsegmentation) v1.0.0rc2 or higher
+
+All the commands below rely on the correct configuration of `PYTHONPATH`, which should point to the project's directory so that Python can locate the module files. In `isnet/` root directory, run the following line to add the current directory to `PYTHONPATH`:
+
+```shell
+export PYTHONPATH=`pwd`:$PYTHONPATH
+```
+
+### Training commands
+
+```shell
+mim train mmsegmentation configs/isnet_r50-d8_8xb2-160k_cityscapes-512x1024.py --work-dir work_dirs/isnet
+```
+
+To train on multiple GPUs, e.g. 8 GPUs, run the following command:
+
+```shell
+mim train mmsegmentation configs/isnet_r50-d8_8xb2-160k_cityscapes-512x1024.py --work-dir work_dirs/isnet --launcher pytorch --gpus 8
+```
+
+### Testing commands
+
+```shell
+mim test mmsegmentation configs/isnet_r50-d8_8xb2-160k_cityscapes-512x1024.py --work-dir work_dirs/isnet --checkpoint ${CHECKPOINT_PATH}
+```
+
+| Method | Backbone | Crop Size | Lr schd | Mem (GB) | Inf time (fps) |  mIoU | mIoU(ms+flip) | config                                                          | download                                                                                                                 |
+| ------ | -------- | --------- | ------: | -------- | -------------- | ----: | ------------: | --------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------ |
+| ISNet  | R-50-D8  | 512x1024  |       - | -        | -              | 79.32 |         80.88 | [config](configs/isnet_r50-d8_8xb2-160k_cityscapes-512x1024.py) | [model](https://download.openmmlab.com/mmsegmentation/v0.5/isnet/isnet_r50-d8_cityscapes-512x1024_20230104-a7a8ccf2.pth) |
+
+## Citation
+
+```bibtex
+@article{Jin2021ISNetII,
+  title={ISNet: Integrate Image-Level and Semantic-Level Context for Semantic Segmentation},
+  author={Zhenchao Jin and B. Liu and Qi Chu and Nenghai Yu},
+  journal={2021 IEEE/CVF International Conference on Computer Vision (ICCV)},
+  year={2021},
+  pages={7169-7178}
+}
+```

projects/isnet/configs/isnet_r50-d8_8xb2-160k_cityscapes-512x1024.py

@@ -0,0 +1,80 @@
+_base_ = [
+    '../../../configs/_base_/datasets/cityscapes.py',
+    '../../../configs/_base_/default_runtime.py',
+    '../../../configs/_base_/schedules/schedule_80k.py'
+]
+
+data_root = '../../data/cityscapes/'
+train_dataloader = dict(dataset=dict(data_root=data_root))
+val_dataloader = dict(dataset=dict(data_root=data_root))
+test_dataloader = dict(dataset=dict(data_root=data_root))
+
+custom_imports = dict(imports=['projects.isnet.decode_heads'])
+
+norm_cfg = dict(type='SyncBN', requires_grad=True)
+data_preprocessor = dict(
+    type='SegDataPreProcessor',
+    mean=[123.675, 116.28, 103.53],
+    std=[58.395, 57.12, 57.375],
+    bgr_to_rgb=True,
+    pad_val=0,
+    seg_pad_val=255)
+
+model = dict(
+    type='EncoderDecoder',
+    data_preprocessor=data_preprocessor,
+    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='ISNetHead',
+        in_channels=(256, 512, 1024, 2048),
+        input_transform='multiple_select',
+        in_index=(0, 1, 2, 3),
+        channels=512,
+        dropout_ratio=0.1,
+        transform_channels=256,
+        concat_input=True,
+        with_shortcut=False,
+        shortcut_in_channels=256,
+        shortcut_feat_channels=48,
+        num_classes=19,
+        norm_cfg=norm_cfg,
+        align_corners=False,
+        loss_decode=[
+            dict(
+                type='CrossEntropyLoss',
+                use_sigmoid=False,
+                loss_weight=1.0,
+                loss_name='loss_o'),
+            dict(
+                type='CrossEntropyLoss',
+                use_sigmoid=False,
+                loss_weight=0.4,
+                loss_name='loss_d'),
+        ]),
+    auxiliary_head=dict(
+        type='FCNHead',
+        in_channels=1024,
+        in_index=2,
+        channels=512,
+        num_convs=1,
+        concat_input=False,
+        dropout_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)),
+    train_cfg=dict(),
+    # test_cfg=dict(mode='slide', crop_size=(769, 769), stride=(513, 513))
+    test_cfg=dict(mode='whole'))

projects/isnet/decode_heads/__init__.py

@@ -0,0 +1,3 @@
+from .isnet_head import ISNetHead
+
+__all__ = ['ISNetHead']