- 1自主系统的未来发展趋势
- 2利用手机热点（手机网络共享）+IPv6+DDNS实现外部网络访问家中电脑等设备_手机热点ipv6
- 3AGI：人工智能大模型领域实战篇—设计一个类似GPT-3.5/GPT-4的大模型从开发→部署→应用需要经过的八大步骤、为什么只有少数公司和机构能够承担这样的训练成本之详细介绍_大模型实战
- 4RK3588上部署yoloV5_rk3588 yolov5
- 5SCDN是什么？
- 6Spark Streaming 编程新手入门指南_spark streaming是核心spark apl 的扩展,支持可扩展,实时数据流的
- 7AI 的架构与核心
- 8利用python中GDAL读写tif文件_python gdal 读取 超大图片
- 9阿里张勇：所有行业都值得用大模型重新做一遍！
- 10[转载]uni-app 换肤实现-原生导航栏、tabbar和页面全部替换_uniapp 用一个组件替换当前页所有内容
人脸分割数据集和论文汇总（face parsing or face segmentation）
赞
踩
最近要研究人脸分割的任务，这里总结一些当前收集到的信息
数据集：人脸分割的data set：
| LFW | 只有头发 | |
| Helen | 五官分割数据，大概2330张 | |
| FASSEG:FAce Smantic SEGmentation repository | 数据有点少 | |
| Mut1ny head/face segmentation dataset | 16557张人脸分割数据集 | |
| LaPa[京东的数据集] | 22176张人脸数据集；数据集是从300W-LP与megaface两个数据集中选取 | |
| iBugMask | 1000张人脸数据集 | |
| CelebAMask-HQ | 包含人脸3W张高清的人像 | |
| EasyPortrait [2023] | 20000张高清图片，非常重要的人像分割数据集；同时给出了大量的预训练人像分割模型 | GitHub - hukenovs/easyportrait: EasyPortrait - Face Parsing and Portrait Segmentation Dataset |
| Short-video face segmentation Dataset[2021] | 下载不到了 | |
| FaceSynthetics[2022] | 微软人脸合成数据集，有10W张合成数据 |
论文集
| Mask-FPAN: Semi-Supervised Face Parsing in the Wild With De-Occlusion and UV GAN | 包括了对遮挡的检测 | |
| Parameter Efficient Local Implicit Image Function Network for Face Segmentation[2023] | Adobe的工作，数据在Lapa和CelebAMask-HQ目前移动端人脸解析最先进的水平了，参数2.29M，FPS 110,所有的检测结果都在85%以上 | |
| On Efficient Real-Time Semantic Segmentation: A Survey[2022] | ||
| Occlusion-Aware Deep Convolutional Neural Network for Face Parsing[2022] | 遮挡 | |
| Detailed feature extraction network-based fine-grained face segmentation[2022] | 韩国的工作，号称能够抓取更多的细节，没有开源代码，论文看结果还不错 | |
| A Masked Self-Supervised Pretraining Method for Face Parsing | 好像灌水的嫌疑很大 | |
| A comprehensive survey on semantic facial attribute editing using generative adversarial networks[2022] | ||
| 3D Face Parsing via Surface Parameterization and 2D Semantic Segmentation Network[2022] | 3D的面部解析 | |
| Decoupled Multi-task Learning with Cyclical Self-Regulation for Face Parsing[2022] | backbone:resnet-101 | https://github.com/deepinsight/insightface/tree/master/parsing/dml_csr |
| RoI Tanh-polar Transformer Network for Face Parsing in the Wild[2021] | backbone:resnet-50 解决姿态变化的问题 | |
| Shuffle Transformer with Feature Alignment for Video Face Parsing[2021] | Short Video Face Parsing Challenge 第一名（腾讯） 1、Net: backbone:Shuffle Transformer ，几个网络一顿串并链接，就OK了 2、loss没讲到 | |
| Edge Aware Network for Face Parsing[2021] | Short Video Face Parsing Challenge 第三名（大连理工） 头部检测，face segmentation 1、头部区域检测Net： backbone: ResNet50， FPN and DCN 检测头部区域 2、face segmentation net：backbond:HRNet_W48+ upsamples:OCRNet | |
| Fake it till you make it: face analysis in the wild using synthetic data alone | 微软合成数据进行人脸解析的训练，在face parsing测试的性能： 1）net:UNet, backbone: ResNet-18 2）loss 是交叉熵损失 3）Label Adaptation自适应网络，没看太懂（需要分析一下数据） | |
| General Facial Representation Learning in a Visual-Linguistic Manner[2021] | 预训练网络600多M | |
| AGRNet: Adaptive Graph Representation Learning and Reasoning for Face Parsing[2021] | backbone:resnet-101 | |
| Edge-aware Graph Representation Learning and Reasoning for Face Parsing[2020] | ||
| Towards Learning Structure via Consensus for Face Segmentation and Parsing[2020] | 看起来只是分割了面部区域和遮挡 | |
| Semantic Segmentation using DeepLabv3[2020] | DeepLabV3 语义做面部语义分割 | https://medium.com/technovators/semantic-segmentation-using-deeplabv3-ce68621e139e |
| Face Segmentation: A Journey From Classical to Deep Learning Paradigm, Approaches, Trends, and Directions[2020] | 2020年的综述 | |
| EHANet：An Effective Hierarchical Aggregation Network for Face Parsing[2020] | 78.19%的miou | |
| HLNet: A Unified Framework for Real-Time Segmentation and Facial Skin Tones Evaluation | 头发、皮肤分割和皮肤颜色估计算法 | |
| A Multi-Task Framework for Facial Attributes Classification through End-to-End Face Parsing and Deep Convolutional Neural Networks[2020] | ||
| Face Segmentation: A Journey From Classical to Deep Learning Paradigm, Approaches, Trends, and Directions[2020] | ||
| End-to-end face parsing via interlinked convolutional neural networks[2020] | ||
| A collection of deep learning frameworks ported to Keras for face analysis.[2019] | Face detection: S3FD model ported from 1adrianb/face-alignment. MTCNN model ported from davidsandberg/facenet. Face landmarks detection: 2DFAN-4, 2DFAN-2, and 2DFAN-1 models ported from 1adrianb/face-alignment. Face parsing: BiSeNet model ported from zllrunning/face-parsing.PyTorch. Eye region landmarks detection: ELG model is ported from swook/GazeML. Face verification: InceptionResNetV1 model (model name: 20180402-114759) ported from davidsandberg/facenet. LResNet100E-IR model ported from deepinsight/insightface. IR50 model ported from ZhaoJ9014/face.evoLVe.PyTorch. Gender and age estimation: MobileNet model ported from deepinsight/insightface. | |
| BiSeNet: Bilateral Segmentation Network for Real-time Semantic Segmentation | 实时的图像分割算法，同时支持了face parsing；有人在手机上查看效果，还不错，网络基于ResNet18 实测 手机上的推理性能大约是：bisenet inference elapse time: 458ms | |
| BiSeNet V2: Bilateral Network with Guided Aggregation for Real-time Semantic Segmentation | 语义分割 | 轻量级实时分割经典BiSeNet及其进化BiSeNet V2_轻量级分割算法_AI算法修炼营的博客-CSDN博客 | |
| Face Parsing With RoI Tanh-Warping[2019] | 微软研究院和厦门大学合作的文章 | GitHub - Eskender-B/roi-tanh: Face Parsing with RoI Tanh-Warping |
| Accurate Facial Image Parsing at Real-Time Speed[2019] | 使用VGG16作为骨干网络 | |
| A high-efficiency framework for constructing large-scale face parsing benchmark[2019] | 京东的一个人脸分割模型，主要提到两点： 1、生成了lapa数据集 2、提到通过一个边界网络(BSPNet)增强人脸分割的精度 3、backbone:ResNet101，并使用imagenet训练的模型的预训练参数 从resnet101来看，基本就不具有实时性 4、Upsamples： spatial pyramid pooling module(SPP) | |
| A New Dataset and Boundary-Attention Semantic Segmentation for Face Parsing[2019] | ||
| Face parsing for mobile AR applications[2018] | 2018的工作，自己设计了一个Moblie-Hourglass网络，参考编解码的结构，并且最后加了一个Dense Blocks来对结果着色。论文基于Helen的数据，没有给出损失函数的设计以及实际分割结果的比较。网络设计异常简单，不知道实际效果如何 | |
| Mobile Real-time Video Segmentation[2018] | google基于U-net，实现了video face parsing，并且达到了98%的 iou；从演示的效果上看，只分割了背景，没有对脸做分割，这两篇文章都说明U-NET能够在移动端达到实时的效果，但是人脸分割的效果如何，还是不知道的 | https://ai.googleblog.com/2018/03/mobile-real-time-video-segmentation.html |
| Real-time Portrait Segmentation on Smartphones[2018] | 2018年基于U-net的portrait 分割工作，不做 face parsing | https://blog.prismalabs.ai/real-time-portrait-segmentation-on-smartphones-39c84f1b9e66 |
| Deep face segmentation in extremely hard conditions[2018] | ||
| Residual Encoder Decoder Network and Adaptive Prior for Face Parsing | ||
| Face segmentation with CNN and CRF[2018] | 根据关键点进行人脸分割 | GitHub - criminalking/face_segmentation: Face segmentation with CNN and CRF |
| On Face Segmentation, Face Swapping, and Face Perception[2018] | GitHub - YuvalNirkin/face_segmentation: Deep face segmentation in extremely hard conditions | |
| Deep face segmentation in extremely hard conditions[2018] | 人脸分割，解决了遮挡的问题 | |
| Real-Time Facial Segmentation and Performance Capture from RGB Input[2016] | 叫面部皮肤分割更为合适一点 1、数据集： 1）LFW和FaceWareHouse（表情数据集） 2）数据增强： perturbations、croppings、occlusion generation、hand compositings、negative samples containing no faces 2、Network，分两个网络：convolutional neural network(face region) + segmentation refinement network(encode(vgg16) + decode) 3、loss：二分类的损失函数 4、optimizer：learning rate:0.01, momentum:0.9,weight decay:0.0005；5W iterations 收敛 fune-tune negative samples：learning rate:0.001；1W收敛 | |
| Fully Convolutional Networks for Semantic Segmentation | FCN 语义分割的鼻祖 |
