Home

Awesome

<h1 align="center">Deep Image Matting: A Comprehensive Survey</h1> <p align="center"> <a href="https://arxiv.org/abs/2304.04672"><img src="https://img.shields.io/badge/arxiv-Paper-brightgreen" ></a> <a href="https://arxiv.org/pdf/2304.04672.pdf"><img src="https://shields.io/badge/-survey-yellow"></a> <a href="https://opensource.org/license/mit/"><img src="https://img.shields.io/badge/license-MIT-blue"></a> </p> <h4 align="center">This is the official repository of the paper <a href="https://arxiv.org/abs/2304.04672">Deep Image Matting: A Comprehensive Survey</a>.</h4> <h5 align="center"><em>Jizhizi Li, Jing Zhang, and Dacheng Tao<sup>1</sup></em></h5> <h6 align="center">1 The University of Sydney, Sydney, Australia</h6> <p align="center"> <a href="#introduction">Introduction</a> | <a href="#preliminary">Preliminary</a> | <a href="#image-matting-methods">Methods</a> | <a href="#image-matting-datasets">Datasets</a> | <a href="#performance-benchmarking">Benchmark</a> | <a href="#statement">Statement</a> </p>

Introduction

Image matting refers to extracting precise alpha matte from natural images, and it plays a critical role in various downstream applications, such as image editing. The emergence of deep learning has revolutionized the field of image matting and given birth to multiple new techniques, including automatic, interactive, and referring image matting. Here we present a comprehensive review of recent advancements in image matting in the era of deep learning by focusing on two fundamental sub-tasks: auxiliary input-based image matting.

Preliminary

Image matting, which refers to the precise extraction of the soft matte from foreground objects in arbitrary images, has been extensively studied for several decades. The process can be described mathematically as below, where I represents the input image, F represents the foreground image, and B represents the background image. The opacity of the pixel in the foreground is denoted by α<sub>i</sub>, which ranges from 0 to 1. We also show the typical input image, ground truth alpha matte and various auxiliary inputs such as trimap, background, coarse map, user clicks, scribbles, and a text description in the following figure. The text description for this image can be the cute smiling brown dog in the middle of the image.

<img src='src/equation.png' width='400px'>

Image Matting Methods

We compile a timeline of the developments in deep learning-based image matting methods as follows.

We also list a summary of image matting methods organized according to the year of publication, the publication venue, input modality, automaticity, matting target, architecture, and availability of the code (with the link). The list of papers is chronologically ordered. Please note that [U] stands for the unofficial implementation of the code.

<table> <thead> <tr> <th>Year</th> <th>Method</th> <th>Pub.</th> <th>Input</th> <th>Auto.</th> <th>Target</th> <th>Arch.</th> <th>Code</th> </tr> </thead> <tbody align='center'> <tr> <th rowspan="2">2016</th> <td><a href="https://link.springer.com/chapter/10.1007/978-3-319-46448-0_6">Deep automatic portrait matting (DAPM)</a></td> <td>ECCV</td> <td>RGB</td> <td>&check;</td> <td>human<a href="https://jiaya.me/projects/automatting/index.html"><img src="https://shields.io/badge/-dataset-orange"></td> <td>Sequential two-step CNN</td> <td>-</td> </tr> <tr> <td><a href="https://link.springer.com/chapter/10.1007/978-3-319-46475-6_39">Natural image matting using deep convolutional neural networks (DCNN)</a></td> <td>ECCV</td> <td>RGB-Coarse</td> <td>&cross;</td> <td>object</td> <td>One-stage CNN</td> <td>-</td> </tr> <tr> <th rowspan="2">2017</th> <td><a href="https://openaccess.thecvf.com/content_cvpr_2017/html/Xu_Deep_Image_Matting_CVPR_2017_paper.html">Deep image matting (DIM)</a></td> <td>CVPR</td> <td>RGB-Trimap</td> <td>&cross;</td> <td>object<a href="https://sites.google.com/view/deepimagematting"><img src="https://shields.io/badge/-dataset-orange"></td> <td>One-stage CNN+Refine</td> <td><a href="https://github.com/foamliu/Deep-Image-Matting-PyTorch">Github[U]<img src="https://img.shields.io/github/stars/foamliu/Deep-Image-Matting-PyTorch.svg?logo=github&label=Stars"></a></td> </tr> <tr> <td><a href="https://dl.acm.org/doi/10.1145/3123266.3123286">Fast deep matting for portrait animation on mobile phone (FDM)</a></td> <td>MM</td> <td>RGB</td> <td>&check;</td> <td>human</td> <td>Sequantial two-step CNN</td> <td>-</td> </tr> <tr> <th rowspan="6">2018</th> <td><a href="https://openaccess.thecvf.com/content_cvpr_2018/html/Chen_TOM-Net_Learning_Transparent_CVPR_2018_paper.html">Tom-Net: Learning transparent object matting from a single image (TOM-Net)</a></td> <td>CVPR</td> <td>RGB</td> <td>&check;</td> <td>trans.<a href="https://guanyingc.github.io/TOM-Net/"><img src="https://shields.io/badge/-dataset-orange"></td> <td>Sequential two-step CNN+Refine</td> <td><a href="https://github.com/guanyingc/TOM-Net">Github<img src="https://img.shields.io/github/stars/guanyingc/TOM-Net.svg?logo=github&label=Stars"></a></td> </tr> <tr> <td><a href="https://www.ijcai.org/proceedings/2018/0139">Deep propagation based image matting (DMPN)</a></td> <td>IJCAI</td> <td>RGB-Trimap</td> <td>&cross;</td> <td>object</td> <td>One-stage CNN</td> <td>-</td> </tr> <tr> <td><a href="http://bmvc2018.org/contents/papers/0915.pdf">Alphagan: Generative adversarial networks for natural image matting (AlphaGAN)</a></td> <td>BMVC</td> <td>RGB-Trimap</td> <td>&cross;</td> <td>object</td> <td>One-stage GAN</td> <td><a href="https://github.com/CDOTAD/AlphaGAN-Matting">Github[U]<img src="https://img.shields.io/github/stars/CDOTAD/AlphaGAN-Matting.svg?logo=github&label=Stars"></a></td> </tr> <tr> <td><a href="https://dl.acm.org/doi/10.1145/3197517.3201275">Semantic soft segmentation (SSS)</a></td> <td>TOG</td> <td>RGB</td> <td>&check;</td> <td>object</td> <td>Sequential two-stage</td> <td><a href="https://github.com/yaksoy/SemanticSoftSegmentation">Github<img src="https://img.shields.io/github/stars/yaksoy/SemanticSoftSegmentation.svg?logo=github&label=Stars"></a></td> </tr> <tr> <td><a href="https://dl.acm.org/doi/pdf/10.1145/3240508.3240610">Semantic human matting (SHM)</a></td> <td>MM</td> <td>RGB</td> <td>&check;</td> <td>human</td> <td>Sequential two-step CNN</td> <td><a href="https://github.com/lizhengwei1992/Semantic_Human_Matting">Github[U]<img src="https://img.shields.io/github/stars/lizhengwei1992/Semantic_Human_Matting.svg?logo=github&label=Stars"></a></td> </tr> <tr> <td><a href="https://papers.nips.cc/paper_files/paper/2018/hash/653ac11ca60b3e021a8c609c7198acfc-Abstract.html">Active matting (ActiveMatting)</a></td> <td>NeurIPS</td> <td>RGB-Click</td> <td>&cross;</td> <td>object</td> <td>One-stage RNN</td> <td>-</td> </tr> <tr> <th rowspan="5">2019</th> <td><a href="https://openaccess.thecvf.com/content_CVPR_2019/html/Zhang_A_Late_Fusion_CNN_for_Digital_Matting_CVPR_2019_paper.html">A late fusion cnn for digital matting (LF)</a></td> <td>CVPR</td> <td>RGB</td> <td>&check;</td> <td>object<a href="https://github.com/yunkezhang/FusionMatting"><img src="https://shields.io/badge/-dataset-orange"></td> <td>Sequential two-stage CNN</td> <td><a href="https://github.com/yunkezhang/FusionMatting">Github<img src="https://img.shields.io/github/stars/yunkezhang/FusionMatting.svg?logo=github&label=Stars"></td> </tr> <tr> <td><a href="https://openaccess.thecvf.com/content_CVPR_2019/html/Tang_Learning-Based_Sampling_for_Natural_Image_Matting_CVPR_2019_paper.html">Learning-based sampling for natural image matting (SampleNet)</a></td> <td>CVPR</td> <td>RGB-Trimap</td> <td>&cross;</td> <td>object</td> <td>Parallel three-stream CNN</td> <td>-</td> </tr> <tr> <td><a href="https://openaccess.thecvf.com/content_ICCV_2019/html/Lu_Indices_Matter_Learning_to_Index_for_Deep_Image_Matting_ICCV_2019_paper.html">Indices matter: Learning to index for deep image matting (IndexNet)</a></td> <td>ICCV</td> <td>RGB-Trimap</td> <td>&cross;</td> <td>object</td> <td>One-stage CNN</td> <td><a href="https://github.com/poppinace/indexnet_matting">Github<img src="https://img.shields.io/github/stars/poppinace/indexnet_matting.svg?logo=github&label=Stars"></a></td> </tr> <tr> <td><a href="https://openaccess.thecvf.com/content_ICCV_2019/html/Cai_Disentangled_Image_Matting_ICCV_2019_paper.html">Disentangled image matting (AdaMatting)</a></td> <td>ICCV</td> <td>RGB-Trimap</td> <td>&cross;</td> <td>object</td> <td>Parallel two-stream CNN+refine</td> <td>-</td> </tr> <tr> <td><a href="https://openaccess.thecvf.com/content_ICCV_2019/html/Hou_Context-Aware_Image_Matting_for_Simultaneous_Foreground_and_Alpha_Estimation_ICCV_2019_paper.html">Context-aware image matting for simultaneous foreground and alpha estimation (Context-Aware)</a></td> <td>ICCV</td> <td>RGB-Trimap</td> <td>&cross;</td> <td>object</td> <td>Two-stream CNN</td> <td><a href="https://github.com/hqqxyy/Context-Aware-Matting">Github<img src="https://img.shields.io/github/stars/hqqxyy/Context-Aware-Matting.svg?logo=github&label=Stars"></a></td> </tr> <tr> <th rowspan="14">2020</th> <td><a href="https://ojs.aaai.org/index.php/AAAI/article/view/6809">Natural image matting via guided contextual attention (GCA)</a></td> <td>AAAI</td> <td>RGB-Trimap</td> <td>&cross;</td> <td>object</td> <td>One-stage CNN</td> <td><a href="https://github.com/Yaoyi-Li/GCA-Matting">Github<img src="https://img.shields.io/github/stars/Yaoyi-Li/GCA-Matting.svg?logo=github&label=Stars"></a></td> </tr> <tr> <td><a href="https://openaccess.thecvf.com/content_CVPR_2020/html/Sengupta_Background_Matting_The_World_Is_Your_Green_Screen_CVPR_2020_paper.html">Background matting: The world is your green screen (BM)</a></td> <td>CVPR</td> <td>RGB-Bg</td> <td>&cross;</td> <td>human</td> <td>Parallel four-stream CNN</td> <td><a href="https://github.com/senguptaumd/Background-Matting">Github<img src="https://img.shields.io/github/stars/senguptaumd/Background-Matting.svg?logo=github&label=Stars"></a></td> </tr> <tr> <td><a href="https://arxiv.org/abs/2004.03249">Hierarchical opacity propagation for image matting (HOP)</a></td> <td>arXiv</td> <td>RGB-Trimap</td> <td>&cross;</td> <td>object</td> <td>Parallel two-stream CNN</td> <td><a href="https://github.com/Yaoyi-Li/HOP-Matting">Github<img src="https://img.shields.io/github/stars/Yaoyi-Li/HOP-Matting.svg?logo=github&label=Stars"></a></td> </tr> <tr> <td><a href="https://openaccess.thecvf.com/content_CVPR_2020/html/Liu_Boosting_Semantic_Human_Matting_With_Coarse_Annotations_CVPR_2020_paper.html">Boosting semantic human matting with coarse annotations (SHMC)</a></td> <td>CVPR</td> <td>RGB</td> <td>&check;</td> <td>human</td> <td>Sequential two-stage CNN</td> <td>-</td> </tr> <tr> <td><a href="https://arxiv.org/abs/2003.07711">F, b, alpha matting (FBA)</a></td> <td>arXiv</td> <td>RGB-Trimap</td> <td>&cross;</td> <td>object</td> <td>One-stage CNN</td> <td><a href="https://github.com/MarcoForte/FBA_Matting">Github<img src="https://img.shields.io/github/stars/MarcoForte/FBA_Matting.svg?logo=github&label=Stars"></a></a></td> </tr> <tr> <td><a href="https://openaccess.thecvf.com/content_CVPR_2020/html/Qiao_Attention-Guided_Hierarchical_Structure_Aggregation_for_Image_Matting_CVPR_2020_paper.html">Attention-guided hierarchical structure aggregation for image matting (HAtt)</a></td> <td>CVPR</td> <td>RGB</td> <td>&check;</td> <td>object<a href="https://github.com/yuhaoliu7456/CVPR2020-HAttMatting"><img src="https://shields.io/badge/-dataset-orange"></td> <td>One-stage CNN</td> <td>-</td> </tr> <tr> <td><a href="https://ojs.aaai.org/index.php/AAAI/article/view/16432">High-resolution deep image matting (HDMatt)</a></td> <td>AAAI</td> <td>RGB-Trimap</td> <td>&cross;</td> <td>object</td> <td>Parallel two-stream CNN</td> <td>-</td> </tr> <tr> <td><a href="https://link.springer.com/article/10.1007/s11263-021-01541-0">Bridging composite and real: towards end-to-end deep image matting (GFM)</a></td> <td>IJCV</td> <td>RGB</td> <td>&check;</td> <td>human, animal<a href="https://github.com/JizhiziLi/gfm#am-2k"><img src="https://shields.io/badge/-dataset-orange"></td> <td>Parallel two-stream CNN</td> <td><a href="https://github.com/JizhiziLi/gfm">Github<img src="https://img.shields.io/github/stars/JizhiziLi/gfm.svg?logo=github&label=Stars"></a></td> </tr> <tr> <td><a href="https://ojs.aaai.org/index.php/AAAI/article/view/19999">Modnet: Real-time trimap-free portrait matting via objective decomposition (MODNet)</a></td> <td>AAAI</td> <td>RGB</td> <td>&check;</td> <td>human<a href="https://github.com/ZHKKKe/MODNet"><img src="https://shields.io/badge/-dataset-orange"></td> <td>Parallel two-stream CNN</td> <td><a href="https://github.com/ZHKKKe/MODNet">Github<img src="https://img.shields.io/github/stars/ZHKKKe/MODNet.svg?logo=github&label=Stars"></a></td> </tr> <tr> <td><a href="https://openaccess.thecvf.com/content/CVPR2021/html/Dai_Learning_Affinity-Aware_Upsampling_for_Deep_Image_Matting_CVPR_2021_paper.html">Learning affinity-aware upsampling for deep image matting(A2U)</a></td> <td>CVPR</td> <td>RGB-Trimap</td> <td>&cross;</td> <td>object</td> <td>One-stage CNN</td> <td><a href="https://github.com/dongdong93/a2u_matting">Github<img src="https://img.shields.io/github/stars/dongdong93/a2u_matting.svg?logo=github&label=Stars"></a></td> </tr> <tr> <td><a href="https://openaccess.thecvf.com/content/CVPR2021/html/Yu_Mask_Guided_Matting_via_Progressive_Refinement_Network_CVPR_2021_paper.html">Mask guided matting via progressive refinement network (MGMatting)</a></td> <td>CVPR</td> <td>RGB-Coarse</td> <td>&cross;</td> <td>human<a href="https://github.com/yucornetto/MGMatting"><img src="https://shields.io/badge/-dataset-orange"></td> <td>One-stage CNN</td> <td><a href="https://github.com/yucornetto/MGMatting">Github<img src="https://img.shields.io/github/stars/yucornetto/MGMatting.svg?logo=github&label=Stars"></a></td> </tr> <tr> <td><a href="https://openaccess.thecvf.com/content/CVPR2021/html/Wei_Improved_Image_Matting_via_Real-Time_User_Clicks_and_Uncertainty_Estimation_CVPR_2021_paper.html">Improved image matting via real-time user clicks and uncertainty estimation (InteractiveMatting)</a></td> <td>CVPR</td> <td>RGB-Click</td> <td>&cross;</td> <td>object</td> <td>Parallel two-stream CNN</td> <td>-</td> </tr> <tr> <td><a href="https://dl.acm.org/doi/10.1145/3408323">Smart scribbles for image matting (SmartScribbles)</a></td> <td>TOMM</td> <td>RGB-Scribble</td> <td>&cross;</td> <td>object</td> <td>One-stage CNN</td> <td>-</td> </tr> <tr> <td><a href="https://openaccess.thecvf.com/content/CVPR2021/html/Lin_Real-Time_High-Resolution_Background_Matting_CVPR_2021_paper.html">Real-Time High-Resolution Background Matting (BMV2)</a></td> <td>CVPR</td> <td>RGB-Bg</td> <td>&cross;</td> <td>human<a href="https://grail.cs.washington.edu/projects/background-matting-v2/#/datasets"><img src="https://shields.io/badge/-dataset-orange"></td> <td>One-stage CNN+refine</td> <td><a href="https://github.com/PeterL1n/BackgroundMattingV2">Github<img src="https://img.shields.io/github/stars/PeterL1n/BackgroundMattingV2.svg?logo=github&label=Stars"></a></td> </tr> <tr> <th rowspan="10">2021</th> <td><a href="https://openaccess.thecvf.com/content/WACV2021/html/Liu_Towards_Enhancing_Fine-Grained_Details_for_Image_Matting_WACV_2021_paper.html">Towards enhancing fine-grained details for image matting (FDMatting)</a></td> <td>WACV</td> <td>RGB-Trimap</td> <td>&cross;</td> <td>object</td> <td>Two-stream CNN</td> <td>-</td> </tr> <tr> <td><a href="https://openaccess.thecvf.com/content/CVPR2021/html/Sun_Semantic_Image_Matting_CVPR_2021_paper.html">Semantic image matting (SIM)</a></td> <td>CVPR</td> <td>RGB-Trimap</td> <td>&cross;</td> <td>object<a href="https://github.com/nowsyn/SIM"><img src="https://shields.io/badge/-dataset-orange"></td> <td>One-stage CNN</td> <td><a href="https://github.com/nowsyn/SIM">Github<img src="https://img.shields.io/github/stars/nowsyn/SIM.svg?logo=github&label=Stars"></a></td> </tr> <tr> <td><a href="https://dl.acm.org/doi/10.1145/3474085.3475512">Privacy-preserving portrait matting (P3M-Net)</a></td> <td>MM</td> <td>RGB</td> <td>&check;</td> <td>human<a href="https://github.com/JizhiziLi/P3M#ppt-setting-and-p3m-10k-dataset"><img src="https://shields.io/badge/-dataset-orange"></td> <td>Parallel two-stream CNN</td> <td><a href="https://github.com/jizhiziLi/p3m">Github<img src="https://img.shields.io/github/stars/jizhiziLi/p3m.svg?logo=github&label=Stars"></a></td> </tr> <tr> <td><a href="https://openaccess.thecvf.com/content/ICCV2021/html/Yu_Cascade_Image_Matting_With_Deformable_Graph_Refinement_ICCV_2021_paper.html">Cascade image matting with deformable graph refinement (CasDGR)</a></td> <td>ICCV</td> <td>RGB</td> <td>&check;</td> <td>object</td> <td>Parallel two-stream CNN</td> <td>-</td> </tr> <tr> <td><a href="https://www.ijcai.org/proceedings/2021/111">Deep Automatic Natural Image Matting (AIM-Net)</a></td> <td>IJCAI</td> <td>RGB</td> <td>&check;</td> <td>object<a href="https://github.com/JizhiziLi/aim#aim-500"><img src="https://shields.io/badge/-dataset-orange"></td> <td>Parallel two-stream CNN</td> <td><a href="https://github.com/JizhiziLi/aim">Github<img src="https://img.shields.io/github/stars/JizhiziLi/aim.svg?logo=github&label=Stars"></a></td> </tr> <tr> <td><a href="https://dl.acm.org/doi/10.1145/3474085.3475203">Long-range feature propagating for natural image matting (LFPNet)</a></td> <td>MM</td> <td>RGB-Trimap</td> <td>&cross;</td> <td>object</td> <td>Parallel two-stream CNN</td> <td><a href="https://github.com/QLYoo/LFPNet">Github<img src="https://img.shields.io/github/stars/QLYoo/LFPNet.svg?logo=github&label=Stars"></a></td> </tr> <tr> <td><a href="https://openaccess.thecvf.com/content/ICCV2021/html/Xu_Virtual_Multi-Modality_Self-Supervised_Foreground_Matting_for_Human-Object_Interaction_ICCV_2021_paper.html">Virtual Multi-Modality Self-Supervised Foreground Matting for Human-Object Interaction (VMFM)</a></td> <td>ICCV</td> <td>RGB</td> <td>&check;</td> <td>human-object</td> <td>Sequential two-stage CNN</td> <td>-</td> </tr> <tr> <td><a href="https://openaccess.thecvf.com/content/ICCV2021/html/Liu_Tripartite_Information_Mining_and_Integration_for_Image_Matting_ICCV_2021_paper.html">Tripartite Information Mining and Integration for Image Matting (TIMI-Net)</a></td> <td>ICCV</td> <td>RGB-Trimap</td> <td>&cross;</td> <td>object<a href="https://github.com/kelisiya/TIMI-Net"><img src="https://shields.io/badge/-dataset-orange"></td> <td>Parallel three-stream CNN</td> <td><a href="https://github.com/kelisiya/TIMI-Net">Github<img src="https://img.shields.io/github/stars/kelisiya/TIMI-Net.svg?logo=github&label=Stars"></a></td> </tr> <tr> <td><a href="https://www.bmvc2021-virtualconference.com/assets/papers/0697.pdf">Deep Image Matting with Flexible Guidance Input (FGI)</a></td> <td>BMVC</td> <td>RGB-Flexible</td> <td>&cross;</td> <td>object</td> <td>One-stage CNN</td> <td><a href="https://github.com/Charch-630/FGI-Matting">Github<img src="https://img.shields.io/github/stars/Charch-630/FGI-Matting.svg?logo=github&label=Stars"></a></td> </tr> <tr> <td><a href="https://www.bmvc2021-virtualconference.com/assets/papers/1642.pdf">Highly efficient natural image matting (HEMatting)</a></td> <td>BMVC</td> <td>RGB</td> <td>&check;</td> <td>object</td> <td>Sequential two-stage CNN</td> <td>-</td> </tr> <tr> <th rowspan="10">2022</th> <td><a href="https://openaccess.thecvf.com/content/CVPR2022/html/Dai_Boosting_Robustness_of_Image_Matting_With_Context_Assembling_and_Strong_CVPR_2022_paper.html">Boosting Robustness of Image Matting With Context Assembling and Strong Data Augmentation (Rmat)</a></td> <td>CVPR</td> <td>RGB-Trimap</td> <td>&cross;</td> <td>object</td> <td>Parallel two-stream CNN/Transformer</td> <td>-</td> </tr> <tr> <td><a href="https://ieeexplore.ieee.org/document/9730784">Deep interactive image matting with feature propagation (DIIM)</a></td> <td>TIP</td> <td>RGB-Click</td> <td>&cross;</td> <td>object</td> <td>One-stage CNN</td> <td>-</td> </tr> <tr> <td><a href="https://ieeexplore.ieee.org/document/9714230">User-Guided Deep Human Image Matting Using Arbitrary Trimaps (UGDMatting)</a></td> <td>TIP</td> <td>RGB-Flexible</td> <td>&cross;</td> <td>human</td> <td>Parallel two-stream CNN</td> <td>-</td> </tr> <tr> <td><a href="https://ieeexplore.ieee.org/document/9733204">Image matting with deep gaussian process (matting-GP)</a></td> <td>TNNLS</td> <td>RGB-Trimap</td> <td>&cross;</td> <td>object</td> <td>One-stage CNN</td> <td>-</td> </tr> <tr> <td><a href="https://arxiv.org/abs/2203.16828">Rethinking portrait matting with privacy preserving (P3M-ViTAE)</a></td> <td>IJCV</td> <td>RGB</td> <td>&check;</td> <td>human<a href="https://github.com/ViTAE-Transformer/P3M-Net#ppt-setting-and-p3m-10k-dataset"><img src="https://shields.io/badge/-dataset-orange"></td> <td>Parallel two stream CNN/Transformer</td> <td><a href="https://github.com/ViTAE-Transformer/P3M-Net">Github<img src="https://img.shields.io/github/stars/ViTAE-Transformer/P3M-Net.svg?logo=github&label=Stars"></a></td> </tr> <tr> <td><a href="https://dl.acm.org/doi/10.1145/3503161.3548036">Situational Perception Guided Image Matting (SPG-IM)</a></td> <td>MM</td> <td>RGB</td> <td>&check;</td> <td>object</td> <td>Sequential two-stage CNN</td> <td>-</td> </tr> <tr> <td><a href="https://openaccess.thecvf.com/content/CVPR2022/html/Sun_Human_Instance_Matting_via_Mutual_Guidance_and_Multi-Instance_Refinement_CVPR_2022_paper.html">Human instance matting via mutual guidance and multi-instance refinement (HIM)</a></td> <td>CVPR</td> <td>RGB</td> <td>&check;</td> <td>human<a href="https://github.com/nowsyn/InstMatt"><img src="https://shields.io/badge/-dataset-orange"></td> <td>Sequential two-stage CNN</td> <td><a href="https://github.com/nowsyn/InstMatt">Github<img src="https://img.shields.io/github/stars/nowsyn/InstMatt.svg?logo=github&label=Stars"></a></td> </tr> <tr> <td><a href="https://openaccess.thecvf.com/content/CVPR2022/html/Park_MatteFormer_Transformer-Based_Image_Matting_via_Prior-Tokens_CVPR_2022_paper.html">MatteFormer: Transformer-Based Image Matting via Prior-Tokens (MatteFormer)</a></td> <td>CVPR</td> <td>RGB-Trimap</td> <td>&cross;</td> <td>object</td> <td>One-stage CNN/Transformer</td> <td><a href="https://github.com/webtoon/matteformer">Github<img src="https://img.shields.io/github/stars/webtoon/matteformer.svg?logo=github&label=Stars"></a></td> </tr> <tr> <td><a href="https://arxiv.org/abs/2206.05149">Referring image matting (RIM)</a></td> <td>CVPR</td> <td>RGB-Language</td> <td>&cross;</td> <td>object<a href="https://github.com/JizhiziLi/RIM#refmatte"><img src="https://shields.io/badge/-dataset-orange"></td> <td>One-stage CNN</td> <td><a href="http://github.com/jizhiziLi/rim">Github<img src="https://img.shields.io/github/stars/JizhiziLi/rim.svg?logo=github&label=Stars"></a></td> </tr> <tr> <td><a href="https://dl.acm.org/doi/abs/10.1007/978-3-031-19818-2_15">TransMatting: Enhancing Transparent Objects Matting with Transformers (TransMatting)</a></td> <td>ECCV</td> <td>RGB-Trimap</td> <td>&cross;</td> <td>trans.<a href="https://github.com/AceCHQ/TransMatting"><img src="https://shields.io/badge/-dataset-orange"></td> <td>One-stage CNN/Transformer</td> <td><a href="https://github.com/AceCHQ/TransMatting">Github<img src="https://img.shields.io/github/stars/AceCHQ/TransMatting.svg?logo=github&label=Stars"></a></td> </tr> </tbody> </table>

Image Matting Datasets

We list a summary of the image matting datasets, categorized as the synthetic image-based benchmark, natural image-based benchmark, and test sets. The datasets are ordered based on their release date and are described in terms of publication venue, naturalness, matting target, resolution, number of training and test samples, and availability (along with their links). It should be noted that the size of the datasets is calculated based on the number of distinguished foregrounds, except for TOM and RefMatte, which have pre-defined composite rules.

NamePub.NaturalTargetResolution#Train#TestPublicity
DIM-481CVPR'17object1298×108343150Link
TOMCVPR'18transparent-178,000876Link
LF-257CVPR'19human553×75622829Link
HATT-646CVPR'20object1573×173159660Link
PhotoMatte13kCVPR'20human-13665--
SIMCVPR'21object2194×195034850Link
Human-2kICCV'21human2112×20752000100Link
Trans-460ECCV'22transparent3766×382041050Link
HIM2kCVPR'22human1823×14241500500Link
RefMatteCVPR'23object1543×1162450002500Link
AlphaMattingCVPR'09object3056×2340278Link
DAPM-2kECCV'16human600×8001700300Link
SHM-35kMM'18human-525111400-
SHMC-10kCVPR'20human-9324125-
P3M-10kMM'21human1349×132194211000Link
AM-2kIJCV'22animal1471×11951800200Link
Multi-Object-1kMM'22human-object-1000200-
UGD-12kTIP'22human356×31712066700Link
PhotoMatte85CVPR'20human2304×3456-85Link
AIM-500IJCAI'21object1397×1260-500Link
RWP-636CVPR'21human1038×1327-636Link
PPM-100AAAI'22human2997×2875-100Link

Performance Benchmarking

We provide a comprehensive evaluation of representative matting methods in the paper. Here, we present some subjective results of auxiliary-based matting methods on alphamatting.com and automatic matting methods on P3M-500-NP.

Statement

If you are interested in our work, please consider citing the following:

@article{li2023deep,
  title={Deep Image Matting: A Comprehensive Survey},
  author={Jizhizi Li and Jing Zhang and Dacheng Tao},
  journal={ArXiv},
  year={2023},
  volume={abs/2304.04672}
}

This project is under the MIT license. For further questions, please contact <strong><i>Jizhizi Li</i></strong> at jili8515@uni.sydney.edu.au.

Relevant Projects

<a href="https://github.com/ViTAE-Transformer/ViTAE-Transformer-Matting"><img src="https://shields.io/badge/-A_list_of_our_works_in_matting-9cf?style=for-the-badge"></a>

</p>

[1] <strong>Deep Automatic Natural Image Matting, IJCAI, 2021</strong> | Paper | Github <br><em>     Jizhizi Li, Jing Zhang, and Dacheng Tao</em>

[2] <strong>Privacy-preserving Portrait Matting, ACM MM, 2021</strong> | Paper | Github <br><em>     Jizhizi Li<sup></sup>, Sihan Ma<sup></sup>, Jing Zhang, Dacheng Tao</em>

[3] <strong>Bridging Composite and Real: Towards End-to-end Deep Image Matting, IJCV, 2022 </strong> | Paper | Github <br><em>     Jizhizi Li<sup></sup>, Jing Zhang<sup></sup>, Stephen J. Maybank, Dacheng Tao</em>

[4] <strong>Referring Image Matting, CVPR, 2023</strong> | Paper | Github <br><em>     Jizhizi Li, Jing Zhang, and Dacheng Tao</em>

[5] <strong>Rethinking Portrait Matting with Privacy Preserving, IJCV, 2023</strong> | Paper | Github <br><em>     Sihan Ma<sup></sup>, Jizhizi Li<sup></sup>, Jing Zhang, He Zhang, Dacheng Tao</em>