Paddleseg

新特性

* CVPR 2021 AutoNUE语义分割赛道 **冠军方案** 开源！
* 全新开源的超轻量级人像分割模型PPSeg，基于自采的大规模半身人像数据训练，适用于 **视频会议** 等半身像场景
* 新增交互式分割应用场景，基于**seed-based SOTA模型RITM** 提供了基于 **人像** 和 **COCO+LVIS** 训练的权重
* 全新发布的交互式分割工具EISeg，可用于快速标注数据
* 新增人像分割领域的经典模型 **PortraitNet**，新增Transformer系列 SOTA模型 **SwinTransformer**
* 优化模型预测的后处理逻辑，提升模型预测精度

问题修复

* [1123](https://github.com/PaddlePaddle/PaddleSeg/pull/1123) 修复模型剪枝时内存不足的问题
* [1100](https://github.com/PaddlePaddle/PaddleSeg/pull/1100) 修复CrossEntropyLoss 使用 weight权重时训练无法收敛的问题
* [1082](https://github.com/PaddlePaddle/PaddleSeg/pull/1082) 修复模型剪枝脚本运行失败的问题
* [1081](https://github.com/PaddlePaddle/PaddleSeg/pull/1081) 修复了Windows系统下预测脚本输出保存路径不正确的问题
* [1078](https://github.com/PaddlePaddle/PaddleSeg/pull/1078) 修复多卡训练模型时DataLoader未设置work_init_fn导致多进程中所使用同样的random seed的问题
* [34c1bbf](https://github.com/PaddlePaddle/PaddleSeg/commit/34c1bbf545cb730678659753864da85b2ffcb4ee)、[#30860e](https://github.com/PaddlePaddle/PaddleSeg/commit/30860e67ddfa67c02078631a8a7cd54da0f44494) 修复DecoupledSegNet和SFNet导出失败的问题

致谢

<p>
<a href="https://github.com/yazheng0307"><img src="https://avatars.githubusercontent.com/u/50820616?v=4" width=75 height=75></a>
<a href="https://github.com/parap1uie-s"><img src="https://avatars.githubusercontent.com/u/23453851?v=4" width=75 height=75></a>
<a href="https://github.com/AlwaysGemini"><img src="https://avatars.githubusercontent.com/u/34996622?v=4" width=75 height=75></a>
<a href="https://github.com/geoyee"><img src="https://avatars.githubusercontent.com/u/71769312?v=4" width=75 height=75></a>
<a href="https://github.com/linhandev"><img src="https://avatars.githubusercontent.com/u/29757093?v=4" width=75 height=75></a>
<a href="https://github.com/txyugood"><img src="https://avatars.githubusercontent.com/u/7403063?v=4" width=75 height=75></a>
</p>

* 非常感谢 [yazheng0307](https://github.com/yazheng0307) 进行了文档的整理工作 [#1075](https://github.com/PaddlePaddle/PaddleSeg/pull/1075)
* 非常感谢 [parap1uie-s](https://github.com/parap1uie-s) 提交了修复代码 [#1078](https://github.com/PaddlePaddle/PaddleSeg/pull/1078)
* 非常感谢 [AlwaysGemini](https://github.com/AlwaysGemini) 提交了修复代码 [#1081](https://github.com/PaddlePaddle/PaddleSeg/pull/1081)、[#1082](https://github.com/PaddlePaddle/PaddleSeg/pull/1082)
* 非常感谢 [geoyee](https://github.com/geoyee)、[linhandev](https://github.com/linhandev)贡献了交互式分割工具[EISeg](./contrib/EISeg)
* 非常感谢 [txyugood](https://github.com/txyugood) 贡献了PortraitNet模型

------

New Features
* CVPR 2021 AutoNUE Semantic Segmentation Track Technical Report is open sourced!
* An ultra-lightweight portrait segmentation model named PPSeg is open sourced, which is training based on large-scale portrait data and suitable for **video conference**
* We provide interactive segmentation application scenarios, based on the **seed-based SOTA model RITM** and also provide weights training on **portrait** and **COCO+LVIS**.
* A newly released interactive segmentation tool **EISeg** can be used to quickly label data
* Added the popular model **PortraitNet** in the field of portrait segmentation, and added the Transformer series SOTA model **SwinTransformer**
* We optimize the post-processing logic of model prediction to improve model prediction accuracy

Bug Fix
* [1123](https://github.com/PaddlePaddle/PaddleSeg/pull/1123) Fix the problem of insufficient memory during model pruning
* [1100](https://github.com/PaddlePaddle/PaddleSeg/pull/1100) Fix the problem that the weighted CrossEntropyLoss cannot converge during training
* [1082](https://github.com/PaddlePaddle/PaddleSeg/pull/1082) Fix the problem that the model pruning script fails to run
* [1081](https://github.com/PaddlePaddle/PaddleSeg/pull/1081) Fix the problem that the save path of the prediction script under Windows system is incorrect
* [1078](https://github.com/PaddlePaddle/PaddleSeg/pull/1078) Fix an issue where the DataLoader did not set work_init_fn when training the model in multi-card, which caused the same random seed to be used in multiple processes
* [34c1bbf](https://github.com/PaddlePaddle/PaddleSeg/commit/34c1bbf545cb730678659753864da85b2ffcb4ee)/[#30860e](https://github.com/PaddlePaddle/PaddleSeg/commit/30860e67ddfa67c02078631a8a7cd54da0f44494) Fix the problem that DecoupledSegNet and SFNet cannot be successfully exported

Thanks


<p>
<a href="https://github.com/yazheng0307"><img src="https://avatars.githubusercontent.com/u/50820616?v=4" width=75 height=75></a>
<a href="https://github.com/parap1uie-s"><img src="https://avatars.githubusercontent.com/u/23453851?v=4" width=75 height=75></a>
<a href="https://github.com/AlwaysGemini"><img src="https://avatars.githubusercontent.com/u/34996622?v=4" width=75 height=75></a>
<a href="https://github.com/geoyee"><img src="https://avatars.githubusercontent.com/u/71769312?v=4" width=75 height=75></a>
<a href="https://github.com/linhandev"><img src="https://avatars.githubusercontent.com/u/29757093?v=4" width=75 height=75></a>
<a href="https://github.com/txyugood"><img src="https://avatars.githubusercontent.com/u/7403063?v=4" width=75 height=75></a>
</p>

* Thank you very much [yazheng0307](https://github.com/yazheng0307) for organizing the documents [#1075](https://github.com/PaddlePaddle/PaddleSeg/pull/1075)
* Thank you very much [parap1uie-s](https://github.com/parap1uie-s) for submitting the repair code [#1078](https://github.com/PaddlePaddle/PaddleSeg/pull/1078)
* Thank you very much [AlwaysGemini](https://github.com/AlwaysGemini) for submitting the repair code [#1081](https://github.com/PaddlePaddle/PaddleSeg/pull/1081)/[#1082](https://github.com/PaddlePaddle/PaddleSeg/pull/1082)
* Thank you very much [txyugood](https://github.com/txyugood) for contributing the PortraitNet model
* Thank you very much [geoyee](https://github.com/geoyee)/[linhandev](https://github.com/linhandev) for contributing the interactive segmentation tool [EISeg](./contrib/EISeg)

新特性
* 语义分割方向新增医疗分割模型UNet3+、轻量级模型SFNet、ShuffleNetV2等模型。
* 全新增加 **全景分割** 场景，支持训练、评估、预测以及可视化等能力，新增Anchor-Free的SOTA模型Panoptic-DeepLab。
* 完善部署能力，新增 **移动端部署** 能力和 **web端部署** 能力，并支持添加后处理算子(argmax/softmax)。
* 高精度的人像分割模型humanseg升级为动态图版，并显著优化边缘锯齿问题。
* 升级学习率配置模块，新增10种学习率策略，涵盖了业界主流学习率调度方式。
* 新增Weighted Cross Entropy Loss、L1 Loss、MSE Loss，适用于不同场景下的模型优化。

Bug修复
* [1016](https://github.com/PaddlePaddle/PaddleSeg/pull/1016) 修复NonLocal2D模块在非gaussian模式下shape不一致的问题。
* [1007](https://github.com/PaddlePaddle/PaddleSeg/pull/1007) 修复RandomRotation和RandomScaleAspect在未传入Label时无法正确调用的问题。
* [1006](https://github.com/PaddlePaddle/PaddleSeg/pull/1006) 修复EMANet无法进行单卡训练的问题。
* [995](https://github.com/PaddlePaddle/PaddleSeg/pull/995) 修复了PaddleSeg在PaddlePaddle 2.1版本中存在的兼容性问题。
* [980](https://github.com/PaddlePaddle/PaddleSeg/pull/980) 修复DecoupledSegNet在PaddlePaddle 2.1版本中训练失败的问题。
* [975](https://github.com/PaddlePaddle/PaddleSeg/pull/975) 修复滑窗预测图像小于窗口大小时无法正确预测的问题。
* [971](https://github.com/PaddlePaddle/PaddleSeg/pull/971) 修复ResizeByLong进行数据增强，在预测时没有正确恢复尺寸的问题。

New Features
* New semantic segmentation models such as the medical segmentation model UNet3+, the lightweight model SFNet, and ShuffleNetV2 have been added.
* Newly added **panoramic segmentation** scenes, supporting training, evaluation, prediction and visualization capabilities, and new Anchor-Free SOTA model Panoptic-DeepLab.
* Improve deployment capabilities, add **mobile deployment** and **web deployment** capabilities, and support the addition of post-processing operators (argmax/softmax).
* The high-precision portrait segmentation model **humanseg** is upgraded to dynamic graph version, and the edge aliasing problem is significantly optimized.
* Upgrade the learning rate configuration module and add 10 new learning rate strategies, covering the mainstream learning rate scheduling methods in the industry.
* Added Weighted Cross Entropy Loss, L1 Loss, and MSE Loss, which are suitable for model optimization in different scenarios.

Bug Fix
* [1016](https://github.com/PaddlePaddle/PaddleSeg/pull/1016) Fix the problem that the shape of NonLocal2D module is inconsistent in non-gaussian mode.
* [1007](https://github.com/PaddlePaddle/PaddleSeg/pull/1007) Fixed an issue where RandomRotation and RandomScaleAspect could not be called correctly when Label was not passed in.
* [1006](https://github.com/PaddlePaddle/PaddleSeg/pull/1006) Fix the problem that EMANet cannot be trained in single card.
* [995](https://github.com/PaddlePaddle/PaddleSeg/pull/995) Fixed the compatibility issue of PaddleSeg in PaddlePaddle 2.1 version.
* [980](https://github.com/PaddlePaddle/PaddleSeg/pull/980) Fixed the problem that DecoupledSegNet failed to train in PaddlePaddle 2.1.
* [975](https://github.com/PaddlePaddle/PaddleSeg/pull/975) Fix the problem that the sliding window prediction image cannot be correctly predicted when the image is smaller than the window size.
* [971](https://github.com/PaddlePaddle/PaddleSeg/pull/971) Fix the problem that ResizeByLong does not restore the size correctly in predict phase.

新特性
* 全新发布2.0版本，全面升级至动态图，支持20+分割模型，4个骨干网络，5个数据集，9种Loss：
* 分割模型：ANN、BiSeNetV2、DANet、DeeplabV3、DeeplabV3+、FCN、FastSCNN、Gated-scnn、GCNet、HarDNet、OCRNet、PSPNet、UNet、UNet++、U²Net、Attention UNet、Decoupled SegNet、EMANet、DNLNet、ISANet
* 骨干网络：ResNet, HRNet, MobileNetV3, Xception
* 数据集：Cityscapes, ADE20K, Pascal VOC, Pascal Context, COCO Stuff
* Loss：CrossEntropy Loss、BootstrappedCrossEntropy Loss、Dice Loss、BCE Loss、OhemCrossEntropyLoss、RelaxBoundaryLoss、OhemEdgeAttentionLoss、Lovasz Hinge Loss、Lovasz Softmax Loss
* 提供基于Cityscapes和Pascal Voc数据集的高质量预训练模型 50+
* 支持多卡GPU并行评估，提供了高效的指标计算功能。支持多尺度评估/翻转评估/滑动窗口评估等多种评估方式。
* 支持XPU模型训练，包括DeepLabv3、HRNet、UNet。
* 开源了基于Hierarchical Multi-Scale Attention结构的语义分割模型，在Cityscapes验证集上达到87% mIoU。
* 动态图模式支持模型在线量化、剪枝等模型压缩功能。
* 动态图下支持模型动转静，实现高性能部署。

New Features
* We newly released version 2.0 which has been fully upgraded to dynamic graphics. It supports more than 20 segmentation models, 4 backbone networks, , 5 datasets and 9 losses:
* Segmentation models: ANN, BiSeNetV2, DANet, DeeplabV3, DeeplabV3+, FCN, FastSCNN, Gated-scnn, GCNet, HarDNet, OCRNet, PSPNet, UNet, UNet++, U²Net, Attention UNet, Decoupled SegNet, EMANet, DNLNet, ISANet
* Backbone networks: ResNet, HRNet, MobileNetV3, and Xception
* Datasets: Cityscapes, ADE20K, Pascal VOC, Pascal Context, COCO Stuff
* Losses: CrossEntropy Loss, BootstrappedCrossEntropy Loss, Dice Loss, BCE Loss, OhemCrossEntropyLoss, RelaxBoundaryLoss, OhemEdgeAttentionLoss, Lovasz Hinge Loss, Lovasz Softmax Loss
* We provide more than 50 high quality pre-trained models based on Cityscapes and Pascal Voc datasets.
* The new version support multi-card GPU parallel evaluation for more efficient metrics calculation. It also support multiple evaluation methods such as multi-scale evaluation/flip evaluation/sliding window evaluation.
* XPU model training including DeepLabv3, HRNet, UNet, is available now.
* We open source a semantic segmentation model based on the Hierarchical Multi-Scale Attention structure, and it reached 87% mIoU on the Cityscapes validation set.
* The dynamic graph mode supports model compression functions such as online quantification and pruning.
* The dynamic graph mode supports model export for high-performance deployment.

新特性
* 全新发布2.0-rc版本，全面升级至动态图，支持15+分割模型，4个骨干网络，3个数据集，4种Loss：
* 分割模型：ANN, BiSeNetV2, DANet, DeeplabV3, DeeplabV3+, FCN, FastSCNN, Gated-scnn, GCNet, HarDNet, OCRNet, PSPNet, UNet, UNet++, U^2Net, Attention UNet
* 骨干网络：ResNet, HRNet, MobileNetV3, Xception
* 数据集：Cityscapes, ADE20K, Pascal VOC
* Loss：CrossEntropy Loss、BootstrappedCrossEntropy Loss、Dice Loss、BCE Loss
* 提供基于Cityscapes和Pascal Voc数据集的高质量预训练模型 40+。
* 支持多卡GPU并行评估，提供了高效的指标计算功能。支持多尺度评估/翻转评估/滑动窗口评估等多种评估方式。


New Features
* Newly release 2.0-rc version, fully upgraded to dynamic graph. It supports 15+ segmentation models, 4 backbone networks, 3 datasets, and 4 types of loss functions:
* Segmentation models: ANN, BiSeNetV2, DANet, DeeplabV3, DeeplabV3+, FCN, FastSCNN, Gated-scnn, GCNet, OCRNet, PSPNet, UNet, and U^2Net
* Backbone networks: ResNet, HRNet, MobileNetV3, and Xception
* Datasets: Cityscapes, ADE20K, and Pascal VOC
* Loss: CrossEntropy Loss、BootstrappedCrossEntropy Loss、Dice Loss、BCE Loss.
* Provide 40+ high quality pre-trained models based on Cityscapes and Pascal Voc datasets.
* Support multi-card GPU parallel evaluation. This provides the efficient index calculation function. Support multiple evaluation methods such as multi-scale evaluation/flip evaluation/sliding window evaluation.

新特性
* 增加多尺度评估/翻转评估/滑动窗口评估等功能。
* 支持多卡GPU并行评估，提供了高效的指标计算功能。
* 增加Pascal VOC 2012数据集。
* 新增在Pascal VOC 2012数据集上的高精度预训练模型，详见[模型库](./dygraph/configs/)。
* 支持对PNG格式的伪彩色图片进行预测可视化。

New Features
* Add multi-scale/flipping/sliding-window inference.
* Add the fast multi-GPUs evaluation, and high-efficient metric calculation.
* Add Pascal VOC 2012 dataset.
* Add high-accuracy pre-trained models on Pascal VOC 2012, see [detailed models](./dygraph/configs/).
* Support visualizing pseudo-color images in PNG format while predicting.

* 全面支持Paddle2.0-rc动态图模式，推出PaddleSeg[动态图体验版](./dygraph/)
* 发布大量动态图模型，支持11个分割模型，4个骨干网络，3个数据集：
* 分割模型：ANN, BiSeNetV2, DANet, DeeplabV3, DeeplabV3+, FCN, FastSCNN, GCNet, OCRNet, PSPNet, UNet
* 骨干网络：ResNet, HRNet, MobileNetV3, Xception
* 数据集：Cityscapes, ADE20K, Pascal VOC
* 提供高精度骨干网络预训练模型以及基于Cityscapes数据集的语义分割[预训练模型](./dygraph/configs/)。Cityscapes精度超过**82%**。