Merge branch 'release/v2.0.0-rc' into develop

Mangomomo · Jan 29, 2021 · b24d3da · b24d3da
2 parents 4cec696 + d7bfdfa
commit b24d3da
Show file tree

Hide file tree

Showing 43 changed files with 1,953 additions and 269 deletions.
diff --git a/README.md b/README.md
@@ -8,6 +8,8 @@ English | [简体中文](README_CN.md)
 ![python version](https://img.shields.io/badge/python-3.6+-orange.svg)
 ![support os](https://img.shields.io/badge/os-linux%2C%20win%2C%20mac-yellow.svg)
 
+<img src="./docs/images/seg_news_icon.png" width="50"/> *[2020-12-18] PaddleSeg has released the v2.0.0-rc version, which supports the dynamic graph by default. The static-graph codes have been moved to [legacy](./legacy). See detailed [release notes](./docs/release_notes.md).*
+
 ![demo](./docs/images/cityscapes.gif)
 
 Welcome to PaddleSeg! PaddleSeg is an end-to-end image segmentation development kit developed based on [PaddlePaddle](https://www.paddlepaddle.org.cn), which covers a large number of high-quality segmentation models in different directions such as *high-performance* and *lightweight*. With the help of modular design, we provide two application methods: *Configuration Drive* and *API Calling*. So one can conveniently complete the entire image segmentation application from training to deployment through configuration calls or API calls.
@@ -41,7 +43,9 @@ Welcome to PaddleSeg! PaddleSeg is an end-to-end image segmentation development
 |[Att U-Net](./configs/attention_unet)|-|-|-|-|
 |[U-Net++](./configs/unet_plusplus)|-|-|-|-|
 |[DecoupledSegNet](./configs/decoupled_segnet)|✔|✔|||
-
+|[EMANet](./configs/emanet)|✔|✔|-|-|
+|[ISANet](./configs/isanet)|✔|✔|-|-|
+|[DNLNet](./configs/dnlnet)|✔|✔|-|-|
 ## Dataset
 
 - [x] Cityscapes

diff --git a/README_CN.md b/README_CN.md
@@ -8,6 +8,8 @@
 ![python version](https://img.shields.io/badge/python-3.6+-orange.svg)
 ![support os](https://img.shields.io/badge/os-linux%2C%20win%2C%20mac-yellow.svg)
 
+<img src="./docs/images/seg_news_icon.png" width="50"/> *[2020-12-18] PaddleSeg发布2.0.0rc版，动态图正式成为主目录。静态图已经被移至[legacy](./legacy)子目录下。更多信息请查看详细[更新日志](./docs/release_notes_cn.md)。*
+
 ![demo](./docs/images/cityscapes.gif)
 
 PaddleSeg是基于飞桨[PaddlePaddle](https://www.paddlepaddle.org.cn)开发的端到端图像分割开发套件，涵盖了**高精度**和**轻量级**等不同方向的大量高质量分割模型。通过模块化的设计，提供了**配置化驱动**和**API调用**两种应用方式，帮助开发者更便捷地完成从训练到部署的全流程图像分割应用。
@@ -41,6 +43,9 @@ PaddleSeg是基于飞桨[PaddlePaddle](https://www.paddlepaddle.org.cn)开发的
 |[Att U-Net](./configs/attention_unet)|-|-|-|-|
 |[U-Net++](./configs/unet_plusplus)|-|-|-|-|
 |[DecoupledSegNet](./configs/decoupled_segnet)|✔|✔|||
+|[EMANet](./configs/emanet)|✔|✔|-|-|
+|[ISANet](./configs/isanet)|✔|✔|-|-|
+|[DNLNet](./configs/dnlnet)|✔|✔|-|-|
 
 ## 数据集
 
@@ -93,6 +98,10 @@ python train.py --config configs/quick_start/bisenet_optic_disc_512x512_1k.yml
 * [API参考](./docs/apis)
 * [添加新组件](./docs/add_new_model.md)
 
+## 联系我们
+* 如果你发现任何PaddleSeg存在的问题或者是建议, 欢迎通过[GitHub Issues](https://github.com/PaddlePaddle/PaddleSeg/issues)给我们提issues。
+* 同时欢迎加入PaddleSeg技术交流群：850378321（QQ群1）或者793114768（QQ群2）。
+
 ## 代码贡献
 
 * 非常感谢[jm12138](https://github.com/jm12138)贡献U<sup>2</sup>-Net模型。

diff --git a/configs/README.md b/configs/README.md
@@ -47,7 +47,7 @@
 > 损失函数
 >  * 参数
 >     * types : 损失函数列表
->       * type : 损失函数类型，目前只支持CrossEntropyLoss
+>       * type : 损失函数类型，所支持值请参考损失函数库
 >     * coef : 对应损失函数列表的系数列表
 
 ----

diff --git a/configs/bisenet/README.md b/configs/bisenet/README.md
@@ -2,12 +2,12 @@
 
 ## Reference
 
-> Yu C, Gao C, Wang J, et al. BiSeNet V2: Bilateral Network with Guided Aggregation for Real-time Semantic Segmentation[J]. arXiv preprint arXiv:2004.02147, 2020.
+> Yu, Changqian, Changxin Gao, Jingbo Wang, Gang Yu, Chunhua Shen, and Nong Sang. "BiSeNet V2: Bilateral Network with Guided Aggregation for Real-time Semantic Segmentation." arXiv preprint arXiv:2004.02147 (2020).
 
 ## Performance
 
 ### Cityscapes
 
 | Model | Backbone | Resolution | Training Iters | mIoU | mIoU (flip) | mIoU (ms+flip) | Links |
 |-|-|-|-|-|-|-|-|
-|BiSeNetv2|-|1024x1024|160000|73.19%|74.19%|74.43%|[model](https://paddleseg.bj.bcebos.com/dygraph/cityscapes/bisenet_cityscapes_1024x1024_160k/model.pdparams) \| [log](https://paddleseg.bj.bcebos.com/dygraph/cityscapes/bisenet_cityscapes_1024x1024_160k/train.log) \| [vdl](https://paddlepaddle.org.cn/paddle/visualdl/service/app?id=3ccfaff613de769eadb76f8379afffa5)|
+|BiSeNetv2|-|1024x1024|160000|73.19%|74.19%|74.43%|[model](https://bj.bcebos.com/paddleseg/dygraph/cityscapes/bisenet_cityscapes_1024x1024_160k/model.pdparams) \| [log](https://bj.bcebos.com/paddleseg/dygraph/cityscapes/bisenet_cityscapes_1024x1024_160k/train.log) \| [vdl](https://paddlepaddle.org.cn/paddle/visualdl/service/app?id=3ccfaff613de769eadb76f8379afffa5)|
diff --git a/configs/danet/README.md b/configs/danet/README.md
@@ -2,18 +2,18 @@
 
 ## Reference
 
-> Fu J, Liu J, Tian H, et al. Dual attention network for scene segmentation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2019: 3146-3154.
+> Fu, Jun, Jing Liu, Haijie Tian, Yong Li, Yongjun Bao, Zhiwei Fang, and Hanqing Lu. "Dual attention network for scene segmentation." In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 3146-3154. 2019.
 
 ## Performance
 
 ### Cityscapes
 
 | Model | Backbone | Resolution | Training Iters | mIoU | mIoU (flip) | mIoU (ms+flip) | Links |
 |-|-|-|-|-|-|-|-|
-|DANet|ResNet50_OS8|1024x512|80000|80.27%|-|-|[model](https://paddleseg.bj.bcebos.com/dygraph/cityscapes/danet_resnet50_os8_cityscapes_1024x512_80k/model.pdparams) \| [log](https://paddleseg.bj.bcebos.com/dygraph/cityscapes/danet_resnet50_os8_cityscapes_1024x512_80k/train.log) \| [vdl](https://paddlepaddle.org.cn/paddle/visualdl/service/app?id=6caecf1222a0cc9124a376284a402cbe)|
+|DANet|ResNet50_OS8|1024x512|80000|80.27%|80.53%|-|[model](https://bj.bcebos.com/paddleseg/dygraph/cityscapes/danet_resnet50_os8_cityscapes_1024x512_80k/model.pdparams) \| [log](https://bj.bcebos.com/paddleseg/dygraph/cityscapes/danet_resnet50_os8_cityscapes_1024x512_80k/train.log) \| [vdl](https://paddlepaddle.org.cn/paddle/visualdl/service/app?id=6caecf1222a0cc9124a376284a402cbe)|
 
 ### Pascal VOC 2012 + Aug
 
 | Model | Backbone | Resolution | Training Iters | mIoU | mIoU (flip) | mIoU (ms+flip) | Links |
 |-|-|-|-|-|-|-|-|
-|DANet|ResNet50_OS8|1024x512|40000|78.55%|-|-|[model](https://paddleseg.bj.bcebos.com/dygraph/cityscapes/danet_resnet50_os8_voc12aug_512x512_40k/model.pdparams) \| [log](https://paddleseg.bj.bcebos.com/dygraph/cityscapes/danet_resnet50_os8_voc12aug_512x512_40k/train.log) \| [vdl](https://paddlepaddle.org.cn/paddle/visualdl/service/app?id=51a403a54302bc81dd5ec0310a6d50ba)|
+|DANet|ResNet50_OS8|512x512|40000|78.55%|78.93%|79.68%|[model](https://bj.bcebos.com/paddleseg/dygraph/pascal_voc12/danet_resnet50_os8_voc12aug_512x512_40k/model.pdparams) \| [log](https://bj.bcebos.com/paddleseg/dygraph/pascal_voc12/danet_resnet50_os8_voc12aug_512x512_40k/train.log) \| [vdl](https://paddlepaddle.org.cn/paddle/visualdl/service/app?id=51a403a54302bc81dd5ec0310a6d50ba)|
diff --git a/configs/danet/danet_resnet50_os8_voc12aug_512x512_40k.yml b/configs/danet/danet_resnet50_os8_voc12aug_512x512_40k.yml
@@ -0,0 +1,18 @@
+_base_: '../_base_/pascal_voc12aug.yml'
+
+model:
+  type: DANet
+  backbone:
+    type: ResNet50_vd
+    output_stride: 8
+    pretrained: https://bj.bcebos.com/paddleseg/dygraph/resnet50_vd_ssld_v2.tar.gz
+  num_classes: 19
+  backbone_indices: [2, 3]
+
+loss:
+  types:
+    - type: CrossEntropyLoss
+    - type: CrossEntropyLoss
+    - type: CrossEntropyLoss
+    - type: CrossEntropyLoss
+  coef: [1, 1, 1, 0.4]
diff --git a/configs/dnlnet/README.md b/configs/dnlnet/README.md
@@ -0,0 +1,23 @@
+# Disentangled Non-Local Neural Networks
+
+## Reference
+
+> Minghao Yin, Zhuliang Yao, Yue Cao, Xiu Li, Zheng Zhang, Stephen Lin, Han Hu:
+Disentangled Non-local Neural Networks. ECCV (15) 2020: 191-207.
+
+## Performance
+
+### Cityscapes
+
+| Model | Backbone | Resolution | Training Iters | mIoU | mIoU (flip) | mIoU (ms+flip) |Links |
+|-|-|-|-|-|-|-|-|
+|DNLNet|ResNet50_OS8|1024x512|80000|79.95%|80.43%|-|[model](https://paddleseg.bj.bcebos.com/dygraph/cityscapes/dnlnet_resnet50_os8_cityscapes_1024x512_80k/model.pdparams) \| [log](https://paddleseg.bj.bcebos.com/dygraph/cityscapes/dnlnet_resnet50_os8_cityscapes_1024x512_80k/train.log) \| [vdl](https://paddlepaddle.org.cn/paddle/visualdl/service/app?id=922cf0682c5e684507ab54a14ef12847)|
+|DNLNet|ResNet101_OS8|1024x512|80000|81.03%|81.38%|-|[model](https://paddleseg.bj.bcebos.com/dygraph/cityscapes/dnlnet_resnet101_os8_cityscapes_1024x512_80k/model.pdparams) \| [log](https://paddleseg.bj.bcebos.com/dygraph/cityscapes/dnlnet_resnet101_os8_cityscapes_1024x512_80k/train.log) \| [vdl](https://paddlepaddle.org.cn/paddle/visualdl/service/app?id=3e0d13c4d9dbf4115bbba2abdc88122c)|
+
+### Pascal VOC 2012 + Aug
+
+| Model | Backbone | Resolution | Training Iters | mIoU | mIoU (flip) | mIoU (ms+flip) | Links |
+|-|-|-|-|-|-|-|-|
+|DNLNet|ResNet50_OS8|512x512|40000|80.89%|81.31%|81.56%|[model](https://paddleseg.bj.bcebos.com/dygraph/pascal_voc12/dnlnet_resnet50_os8_voc12aug_512x512_40k/model.pdparams) \| [log](https://paddleseg.bj.bcebos.com/dygraph/pascal_voc12/dnlnet_resnet50_os8_voc12aug_512x512_40k/train.log) \| [vdl](https://paddlepaddle.org.cn/paddle/visualdl/service/app?id=8877c77bef8b227af22c5eb3017138ce)|
+|DNLNet|ResNet101_OS8|512x512|40000|80.49%|80.83%| 81.33%|[model](https://paddleseg.bj.bcebos.com/dygraph/pascal_voc12/dnlnet_resnet101_os8_voc12aug_512x512_40k/model.pdparams) \| [log](https://paddleseg.bj.bcebos.com/dygraph/pascal_voc12/dnlnet_resnet101_os8_voc12aug_512x512_40k/train.log) \| [vdl](https://paddlepaddle.org.cn/paddle/visualdl/service/app?id=1d42c22da1c465d9a38e4204bebeeb54)|
+
diff --git a/configs/dnlnet/dnlnet_resnet101_os8_cityscape_1024x512_80k.yml b/configs/dnlnet/dnlnet_resnet101_os8_cityscape_1024x512_80k.yml
@@ -0,0 +1,31 @@
+_base_: '../_base_/cityscapes.yml'
+
+batch_size: 2
+iters: 80000
+
+model:
+  type: DNLNet
+  backbone:
+    type: ResNet101_vd
+    output_stride: 8
+    pretrained: https://bj.bcebos.com/paddleseg/dygraph/resnet101_vd_ssld.tar.gz
+  num_classes: 19
+
+optimizer:
+  type: sgd
+  momentum: 0.9
+  weight_decay: 0.00004
+
+learning_rate:
+  value: 0.01
+  decay:
+    type: poly
+    power: 0.9
+
+
+loss:
+  types:
+    - type: CrossEntropyLoss
+    - type: CrossEntropyLoss
+  coef: [1, 0.4]
+
diff --git a/configs/dnlnet/dnlnet_resnet101_os8_voc12aug_512x512_40k.yml b/configs/dnlnet/dnlnet_resnet101_os8_voc12aug_512x512_40k.yml
@@ -0,0 +1,25 @@
+_base_: '../_base_/pascal_voc12aug.yml'
+
+model:
+  type: DNLNet
+  backbone:
+    type: ResNet101_vd
+    output_stride: 8
+    pretrained: https://bj.bcebos.com/paddleseg/dygraph/resnet101_vd_ssld.tar.gz
+
+optimizer:
+  type: sgd
+  momentum: 0.9
+  weight_decay: 4.0e-05
+
+learning_rate:
+  value: 0.01
+  decay:
+    type: poly
+    power: 0.9
+
+loss:
+  types:
+    - type: CrossEntropyLoss
+    - type: CrossEntropyLoss
+  coef: [1, 0.4]
diff --git a/configs/dnlnet/dnlnet_resnet50_os8_cityscape_1024x512_80k.yml b/configs/dnlnet/dnlnet_resnet50_os8_cityscape_1024x512_80k.yml
@@ -0,0 +1,30 @@
+_base_: '../_base_/cityscapes.yml'
+
+batch_size: 2
+iters: 80000
+
+model:
+  type: DNLNet
+  backbone:
+    type: ResNet50_vd
+    output_stride: 8
+    pretrained: https://bj.bcebos.com/paddleseg/dygraph/resnet50_vd_ssld_v2.tar.gz
+  num_classes: 19
+
+optimizer:
+  type: sgd
+  momentum: 0.9
+  weight_decay: 0.00004
+
+learning_rate:
+  value: 0.01
+  decay:
+    type: poly
+    power: 0.9
+
+
+loss:
+  types:
+    - type: CrossEntropyLoss
+    - type: CrossEntropyLoss
+  coef: [1, 0.4]
diff --git a/configs/dnlnet/dnlnet_resnet50_os8_voc12aug_512x512_40k.yml b/configs/dnlnet/dnlnet_resnet50_os8_voc12aug_512x512_40k.yml
@@ -0,0 +1,25 @@
+_base_: '../_base_/pascal_voc12aug.yml'
+
+model:
+  type: DNLNet
+  backbone:
+    type: ResNet50_vd
+    output_stride: 8
+    pretrained: https://bj.bcebos.com/paddleseg/dygraph/resnet50_vd_ssld_v2.tar.gz
+
+optimizer:
+  type: sgd
+  momentum: 0.9
+  weight_decay: 4.0e-05
+
+learning_rate:
+  value: 0.01
+  decay:
+    type: poly
+    power: 0.9
+
+loss:
+  types:
+    - type: CrossEntropyLoss
+    - type: CrossEntropyLoss
+  coef: [1, 0.4]
diff --git a/configs/emanet/README.md b/configs/emanet/README.md
@@ -0,0 +1,22 @@
+# Expectation-Maximization Attention Networks for Semantic Segmentation
+
+## Reference
+
+> Xia Li, Zhisheng Zhong, Jianlong Wu, Yibo Yang, Zhouchen Lin, Hong Liu:
+Expectation-Maximization Attention Networks for Semantic Segmentation. ICCV 2019: 9166-9175.
+
+## Performance
+
+### Cityscapes
+
+| Model | Backbone | Resolution | Training Iters | mIoU | mIoU (flip) | mIoU (ms+flip) |Links |
+|-|-|-|-|-|-|-|-|
+|EMANet|ResNet50_OS8|1024x512|80000|77.64%|77.98%|78.23%|[model](https://bj.bcebos.com/paddleseg/dygraph/cityscapes/emanet_resnet50_os8_cityscapes_1024x512_80k/model.pdparams) \| [log](https://bj.bcebos.com/paddleseg/dygraph/cityscapes/emanet_resnet50_os8_cityscapes_1024x512_80k/train.log) \| [vdl](https://paddlepaddle.org.cn/paddle/visualdl/service/app?id=3e053a214d60822d6e65445b8614d052)|
+|EMANet|ResNet101_OS8|1024x512|80000|79.41%|79.83%|80.33%|[model](https://bj.bcebos.com/paddleseg/dygraph/cityscapes/emanet_resnet101_os8_cityscapes_1024x512_80k/model.pdparams) \| [log](https://bj.bcebos.com/paddleseg/dygraph/cityscapes/emanet_resnet101_os8_cityscapes_1024x512_80k/train.log) \| [vdl](https://paddlepaddle.org.cn/paddle/visualdl/service/app?id=87be6389cdada711f5c6ada21d9ef6cd)|
+
+### Pascal VOC 2012 + Aug
+
+| Model | Backbone | Resolution | Training Iters | mIoU | mIoU (flip) | mIoU (ms+flip) | Links |
+|-|-|-|-|-|-|-|-|
+|EMANet|ResNet50_OS8|512x512|40000|78.79%|78.90%|79.17%|[model](https://bj.bcebos.com/paddleseg/dygraph/pascal_voc12/emanet_resnet50_os8_voc12aug_512x512_40k/model.pdparams) \| [log](https://bj.bcebos.com/paddleseg/dygraph/pascal_voc12/emanet_resnet50_os8_voc12aug_512x512_40k/train.log) \| [vdl](https://paddlepaddle.org.cn/paddle/visualdl/service/app?id=3e60b80b984a71f3d2b83b8a746a819c)|
+|EMANet|ResNet101_OS8|512x512|40000|79.73%|79.97%| 80.67%|[model](https://bj.bcebos.com/paddleseg/dygraph/pascal_voc12/emanet_resnet101_os8_voc12aug_512x512_40k/model.pdparams) \| [log](https://bj.bcebos.com/paddleseg/dygraph/pascal_voc12/emanet_resnet101_os8_voc12aug_512x512_40k/train.log) \| [vdl](https://paddlepaddle.org.cn/paddle/visualdl/service/app?id=f33479772409766dbc40b5f031cbdb1a)|
diff --git a/configs/emanet/emanet_resnet101_os8_cityscapes_1024x512_80k.yml b/configs/emanet/emanet_resnet101_os8_cityscapes_1024x512_80k.yml
@@ -0,0 +1,31 @@
+_base_: '../_base_/cityscapes.yml'
+
+batch_size: 2
+iters: 80000
+
+model:
+  type: EMANet
+  backbone:
+    type: ResNet101_vd
+    output_stride: 8
+    pretrained: https://bj.bcebos.com/paddleseg/dygraph/resnet101_vd_ssld.tar.gz
+  num_classes: 19
+  ema_channels: 512
+  gc_channels: 256
+  num_bases: 64
+  stage_num: 3
+  momentum: 0.1
+  concat_input: True
+  enable_auxiliary_loss: True
+  align_corners: False
+
+optimizer:
+  type: sgd
+  momentum: 0.9
+  weight_decay: 0.0005
+
+loss:
+  types:
+    - type: CrossEntropyLoss
+    - type: CrossEntropyLoss
+  coef: [1, 0.4]
diff --git a/configs/emanet/emanet_resnet101_os8_voc12aug_512x512_40k.yml b/configs/emanet/emanet_resnet101_os8_voc12aug_512x512_40k.yml
@@ -0,0 +1,28 @@
+_base_: '../_base_/pascal_voc12aug.yml'
+
+model:
+  type: EMANet
+  backbone:
+    type: ResNet101_vd
+    output_stride: 8
+    pretrained: https://bj.bcebos.com/paddleseg/dygraph/resnet101_vd_ssld.tar.gz
+  ema_channels: 512
+  gc_channels: 256
+  num_bases: 64
+  stage_num: 3
+  momentum: 0.1
+  concat_input: True
+  enable_auxiliary_loss: True
+  align_corners: True
+
+optimizer:
+  type: sgd
+  momentum: 0.9
+  weight_decay: 0.0005
+
+
+loss:
+  types:
+    - type: CrossEntropyLoss
+    - type: CrossEntropyLoss
+  coef: [1, 0.4]
diff --git a/configs/emanet/emanet_resnet50_os8_cityscapes_1024x512_80k.yml b/configs/emanet/emanet_resnet50_os8_cityscapes_1024x512_80k.yml
@@ -0,0 +1,32 @@
+_base_: '../_base_/cityscapes.yml'
+
+batch_size: 2
+iters: 80000
+
+model:
+  type: EMANet
+  backbone:
+    type: ResNet50_vd
+    output_stride: 8
+    pretrained: https://bj.bcebos.com/paddleseg/dygraph/resnet50_vd_ssld_v2.tar.gz
+  num_classes: 19
+  ema_channels: 512
+  gc_channels: 256
+  num_bases: 64
+  stage_num: 3
+  momentum: 0.1
+  concat_input: True
+  enable_auxiliary_loss: True
+  align_corners: False
+
+optimizer:
+  type: sgd
+  momentum: 0.9
+  weight_decay: 0.0005
+
+
+loss:
+  types:
+    - type: CrossEntropyLoss
+    - type: CrossEntropyLoss
+  coef: [1, 0.4]