自然图像抠图/视频抠像技术梳理(image matting, video matting)
2013年及之前
https://blog.csdn.net/kezunhai/article/details/41804805
https://www.twblogs.net/a/5b7acdb72b7177392c96d804/zh-cn
Semantic Soft Segmentation
matting算法的典型输入是trimap
https://blog.csdn.net/u011681952/article/details/87862784
image matting benchmask网站
http://alphamatting.com/eval_25.php
video matting benchmask网站
http://videomatting.com/
代码
https://github.com/topics/matting
代码
http://blog.leanote.com/post/610167078@qq.com/Image-Matting
fatal error: opencv2/objdetect.hpp
https://github.com/aromazyl/auto-portrait-matting
输入二值图
参考Vikas Gupta and Shanmuganathan Raman. (2017). "Automatic Trimap Generation for Image Matting"
https://github.com/lnugraha/trimap_generator
自动移除背景的网站
https://www.remove.bg/
基于显著性的图像分割
https://cloud.tencent.com/developer/article/1422621
https://zhuanlan.zhihu.com/p/63246385
在给定二进制掩码的情况下生成trimap
https://xbuba.com/questions/50191862
2017-CVPR-Adobe-Deep Image Matting
2017-CVPR-Adobe-Deep Image Matting
只对人物有比较好的抠图效果
输入还需要一个trimap三元图,感觉可行性不高
https://github.com/Joker316701882/Deep-Image-Matting
2018-SIGGRAPH-MIT-瑞士苏黎世高工-Semantic Soft Segmentation
2018-SIGGRAPH-MIT-瑞士苏黎世联邦理工学院-Semantic Soft Segmentation
(SSS,语义软分割)只有对图像内容的分割,自动抠图
评论:
请问您在本机上运行的效果同论文中的一样好吗?(我用论文中的几个原图做sss,也不如论文给的效果图,尤其是主体并不是同一个颜色)
用到matlab,实际用起来较困难
https://zhuanlan.zhihu.com/p/63211300
MIT提出精细到头发丝的语义分割技术,打造效果惊艳的特效电影
MIT CSAIL 的研究人员开发了一种精细程度远超传统语义分割方法的「语义软分割」技术,连头发都能清晰地在分割掩码中呈现。
https://www.jiqizhixin.com/articles/2018-08-22-8
普通语义分割将每个像素分配到一个类,语义软分割中则每个像素有可能分配到多个类,且目标之间的过渡区域平滑,这对于图像编辑是非常重要的。以往这需要专业的PS人员处理,而本文通过扩展传统的语义分割,将这个过程实现自动化。
https://blog.csdn.net/jacke121/article/details/81708201
其它分割代码
https://github.com/CSAILVision/semantic-segmentation-pytorch
与自然图像消光方法相反,我们依赖于自动生成的语义特征来定义我们的软段而不是trimap,并生成多个软段而不是前景分割。虽然它们看似相似,但自然消光和软分割具有根本差异。以trimap作为输入的自然消光成为前景和背景颜色建模的问题,可能是通过选择颜色样本或传播颜色信息。同时,软分段侧重于检测最适合目标应用的软转换,在我们的例子中是对应于语义边界的软转换。
http://www.pianshen.com/article/5768103992/
2016-ECCV-deep automatic portrait matting
希望通过CNN实现人像自动抠图
数据集的制作是获取Flickr上的图片,根据图片质量进行筛选,裁剪到图片中心基本都是人像。利用之前的closed-form matting和KNN matting方法对图像进行标注,同时也是用Photoshop进行细节修复。
https://blog.csdn.net/qq_36165459/article/details/78550208
alphamatting benchmask
2018-BMVC-AlphaGAN
AlphaGAN: Generative adversarial networks for natural image matting
The trimap are generated while the data are loaded.
This project is an unofficial implementation of AlphaGAN: Generative adversarial networks for natural image matting published at the BMVC 2018. As for now, the result of my experiment is not as good as the paper's.
This is a course project of mine and there may exists some mistakes in this project.
The result is bad. As it was just a course project, I have no longer worked on it. I strongly recommend you to train it yourself.
pytorch
https://github.com/CDOTAD/AlphaGAN-Matting
https://paperswithcode.com/paper/alphagan-generative-adversarial-networks-for
另一个代码
Project aims at generating a large training dataset for alpha matting, training the model and converting the model to ONNX to
be used with OpenCV's DNN module and to integrate some of the best computer vision based best alpha matting algorithms into
OpenCV.
OpenCV's DNN module
https://github.com/Nerdyvedi/GSOC-Opencv-matting
针对MNIST数据集
https://github.com/yjchoe/alphaGAN
代码
https://v-sense.scss.tcd.ie/research/deep-learning/alphagan-generative-adversarial-networks-for-natural-image-matting/
ALPHAGAN:用于自然图像消光的生成对抗网络
特别适用于精细结构的头发,这在实际的消光应用中是非常重要的,例如在电影/电视制作中。
http://www.tuan18.org/archiver/?tid-9269.html
2019-ICCV-Adobe-IndexNet Matting
Indices Matter: Learning to Index for Deep Image Matting
需要输入trimap
PyTorch
https://github.com/poppinace/indexnet_matting
2019-ICCV-Context-aware Matting
Context-aware Image Matting for Simultaneous Foreground and Alpha Estimation
需要输入trimap
TensorFlow
https://github.com/hqqxyy/Context-Aware-Matting
2017-CVPR-瑞士苏黎世高工-Information-Flow matting
《Designing Effective Inter-Pixel Information Flow for Natural Image Matting》
matlab代码
https://github.com/yaksoy/AffinityBasedMattingToolbox
需要输入trimap
c++实现
https://github.com/Nerdyvedi/Information_flow_matting