{"id":20305563,"url":"https://github.com/vitae-transformer/p3m-net","last_synced_at":"2025-04-11T14:51:20.971Z","repository":{"id":177240105,"uuid":"621045792","full_name":"ViTAE-Transformer/P3M-Net","owner":"ViTAE-Transformer","description":"The official repo for [IJCV'23] \"Rethinking Portrait Matting with Privacy Preserving\"","archived":false,"fork":false,"pushed_at":"2024-03-30T18:11:25.000Z","size":58940,"stargazers_count":101,"open_issues_count":6,"forks_count":9,"subscribers_count":4,"default_branch":"main","last_synced_at":"2025-03-25T11:01:40.621Z","etag":null,"topics":["deep-learning","image-matting","pytorch","vision-transformer"],"latest_commit_sha":null,"homepage":"","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/ViTAE-Transformer.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null}},"created_at":"2023-03-29T22:10:18.000Z","updated_at":"2025-03-24T03:52:49.000Z","dependencies_parsed_at":null,"dependency_job_id":"fc025d9c-be2f-4753-9a8c-c42865994a7e","html_url":"https://github.com/ViTAE-Transformer/P3M-Net","commit_stats":null,"previous_names":["vitae-transformer/p3m-net"],"tags_count":0,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ViTAE-Transformer%2FP3M-Net","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ViTAE-Transformer%2FP3M-Net/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ViTAE-Transformer%2FP3M-Net/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ViTAE-Transformer%2FP3M-Net/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/ViTAE-Transformer","download_url":"https://codeload.github.com/ViTAE-Transformer/P3M-Net/tar.gz/refs/heads/main","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":248424717,"owners_count":21101220,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["deep-learning","image-matting","pytorch","vision-transformer"],"created_at":"2024-11-14T17:08:57.481Z","updated_at":"2025-04-11T14:51:20.941Z","avatar_url":"https://github.com/ViTAE-Transformer.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"\u003ch1 align=\"center\"\u003eRethinking Portrait Matting with Privacy Preserving [IJCV-2023]\u003c/h1\u003e\n\n\u003cp align=\"center\"\u003e\n\u003ca href=\"https://arxiv.org/abs/2203.16828\"\u003e\u003cimg  src=\"https://img.shields.io/badge/arXiv-Paper-\u003cCOLOR\u003e.svg\" \u003e\u003c/a\u003e\n\u003ca href=\"https://colab.research.google.com/drive/1pD_XKx31Lgd7zwq46dRpz2jGdsH1ZIay?usp=sharing\"\u003e\u003cimg src=\"https://colab.research.google.com/assets/colab-badge.svg\"\u003e\u003c/a\u003e\n\u003ca href=\"https://opensource.org/license/mit/\"\u003e\u003cimg  src=\"https://img.shields.io/badge/license-MIT-blue\"\u003e\u003c/a\u003e\n\u003ca href=\"hhttps://paperswithcode.com/sota/image-matting-on-p3m-10k?p=rethinking-portrait-matting-with-privacy\"\u003e\u003cimg src=\"https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/rethinking-portrait-matting-with-privacy/image-matting-on-p3m-10k\"\u003e\u003c/a\u003e\n\u003c/p\u003e\n\n\n\u003ch4 align=\"center\"\u003eThis is the official repository of the paper \u003ca href=\"https://arxiv.org/abs/2203.16828\"\u003e[IJCV'23] Rethinking Portrait Matting with Privacy Preserving\u003c/a\u003e. \n\nFor further questions, please contact \u003cstrong\u003e\u003ci\u003eSihan Ma\u003c/i\u003e\u003c/strong\u003e at [sima7436@uni.sydney.edu.au](mailto:sima7436@uni.sydney.edu.au) or \u003cstrong\u003e\u003ci\u003eJizhizi Li\u003c/i\u003e\u003c/strong\u003e at [jili8515@uni.sydney.edu.au](mailto:jili8515@uni.sydney.edu.au).\u003c/h4\u003e\n\n\n\u003ch5 align=\"center\"\u003e\u003cem\u003eSihan Ma\u003csup\u003e\u0026#8727;\u003c/sup\u003e, Jizhizi Li\u003csup\u003e\u0026#8727;\u003c/sup\u003e, Jing Zhang, He Zhang, and Dacheng Tao\u003c/em\u003e\u003c/h5\u003e\n\n\u003cp align=\"center\"\u003e\n  \u003ca href=\"#introduction\"\u003eIntroduction\u003c/a\u003e |\n  \u003ca href=\"#ppt-setting-and-p3m-10k-dataset\"\u003ePPT and P3M-10k\u003c/a\u003e |\n  \u003ca href=\"#p3m-net-and-variants\"\u003eP3M-Net\u003c/a\u003e |\n  \u003ca href=\"#p3m-cp\"\u003eP3M-CP\u003c/a\u003e |\n  \u003ca href=\"#results\"\u003eResults\u003c/a\u003e |\n  \u003ca href=\"#quick-start---train\"\u003eTrain\u003c/a\u003e |\n  \u003ca href=\"#inference-code---how-to-test-on-your-images\"\u003eInference code\u003c/a\u003e |\n  \u003ca href=\"#statement\"\u003eStatement\u003c/a\u003e\n\u003c/p\u003e\n\n\u003cimg src=\"demo/gif/p_3cf7997c.gif\" width=\"25%\"\u003e\u003cimg src=\"demo/gif/2.gif\" width=\"25%\"\u003e\u003cimg src=\"demo/gif/3.gif\" width=\"25%\"\u003e\u003cimg src=\"demo/gif/4.gif\" width=\"25%\"\u003e\n\n***\n\u003e\u003ch3\u003e\u003cstrong\u003e\u003ci\u003e:postbox: News\u003c/i\u003e\u003c/strong\u003e\u003c/h3\u003e\n\u003e\n\u003e [2024-3-31]: [Code for training](./core/README.md) is available!\n\u003e\n\u003e [2023-11-05]: Publish the ViTAE-S and SWIN-T backbone models pretrained on ImageNet that can be used to train our P3M-Net from scratch.\n\u003e\n\u003e [2023-03-28]: The paper has been accepted by the International Journal of Computer Vision ([IJCV](https://www.springer.com/journal/11263))! 🎉\n\u003e\n\u003e [2022-03-31]: Publish the \u003ca href=\"#inference-code---how-to-test-on-your-images\"\u003einference code\u003c/a\u003e and the pretrained model that can be used to test with our SOTA model \u003cstrong\u003eP3M-Net(ViTAE-S)\u003c/strong\u003e on your own privacy-preserving or normal portrait images.\n\u003e \n\u003e [2021-12-06]: Publish the [\u003cstrong\u003eP3M-10k\u003c/strong\u003e](#ppt-setting-and-p3m-10k-dataset) dataset.\n\u003e\n\u003e [2021-11-21]: Publish the conference paper ACM MM 2021 \"[Privacy-preserving Portrait Matting](https://dl.acm.org/doi/10.1145/3474085.3475512)\". The code and data are available at [github repo](https://github.com/JizhiziLi/P3M).\n\u003e\n\u003e Other applications of ViTAE Transformer include: [image classification](https://github.com/ViTAE-Transformer/ViTAE-Transformer/tree/main/Image-Classification) | [object detection](https://github.com/ViTAE-Transformer/ViTAE-Transformer/tree/main/Object-Detection) | [semantic segmentation](https://github.com/ViTAE-Transformer/ViTAE-Transformer/tree/main/Semantic-Segmentation) | [animal pose segmentation](https://github.com/ViTAE-Transformer/ViTAE-Transformer/tree/main/Animal-Pose-Estimation) | [remote sensing](https://github.com/ViTAE-Transformer/ViTAE-Transformer-Remote-Sensing)\n\n\n## Introduction\n\n\n\u003cp align=\"justify\"\u003eRecently, there has been an increasing concern about the privacy issue raised by using personally identifiable information in machine learning. However, previous portrait matting methods were all based on identifiable portrait images.\u003c/p\u003e\n\n\u003cp align=\"justify\"\u003eTo fill the gap, we present \u003ca href=\"#ppt-setting-and-p3m-10k-dataset\"\u003e\u003cstrong\u003eP3M-10k\u003c/strong\u003e\u003c/a\u003e in this paper, which is the first large-scale anonymized benchmark for Privacy-Preserving Portrait Matting. \u003cstrong\u003eP3M-10k\u003c/strong\u003e consists of 10,000 high-resolution face-blurred portrait images along with high-quality alpha mattes. We systematically evaluate both trimap-free and trimap-based matting methods on P3M-10k and find that existing matting methods show different generalization capabilities when following the Privacy-Preserving Training (PPT) setting, 𝑖.𝑒., \u003ci\u003etraining on face-blurred images and testing on arbitrary images\u003c/i\u003e.\u003c/p\u003e\n\n\u003cp align=\"justify\"\u003eTo devise a better trimap-free portrait matting model, we propose \u003ca href=\"#p3m-net\"\u003e\u003cstrong\u003eP3M-Net\u003c/strong\u003e\u003c/a\u003e, consisting of three carefully designed integration modules that can perform privacy-insensitive semantic perception and detail-reserved matting simultaneously. We further design multiple variants of P3MNet with different CNN and transformer backbones and identify the difference of their generalization abilities.\u003c/p\u003e\n\n\u003cp align=\"justify\"\u003eTo further mitigate this issue, we devise a simple yet effective Copy and Paste strategy (P3M-CP) that can borrow facial information from public celebrity images without privacy concerns and direct the network to reacquire the face context at both data and feature level. P3M-CP only brings a few additional computations during training, while enabling the matting model to process both face-blurred and normal images without extra effort during inference.\u003c/p\u003e\n\n\u003cp align=\"justify\"\u003eExtensive experiments on P3M-10k demonstrate the superiority of P3M-Net over state-of-the-art methods and the effectiveness of P3MCP in improving the generalization ability of P3M-Net, implying a great significance of P3M for future research and real-world applications.\u003c/p\u003e\n\n\n## PPT Setting and P3M-10k Dataset\n\n\n\u003cp align=\"justify\"\u003e\u003cstrong\u003ePPT Setting\u003c/strong\u003e: Due to the privacy concern, we propose the \u003cstrong\u003eP\u003c/strong\u003erivacy-\u003cstrong\u003eP\u003c/strong\u003ereserving \u003cstrong\u003eT\u003c/strong\u003eraining (\u003cstrong\u003ePPT\u003c/strong\u003e) setting in portrait matting, 𝑖.𝑒., training on privacy-preserved images (𝑒.𝑔., processed by face obfuscation) and testing on arbitraty images with or without privacy content. As an initial step towards privacy-preserving portrait matting problem, we only define the \u003ci\u003eidentifiable faces\u003c/i\u003e in frontal and some profile portrait images as the private content in this work. \u003c/p\u003e\n\n\n\u003cp align=\"justify\"\u003e\u003cstrong\u003eP3M-10k Dataset\u003c/strong\u003e: To further explore the effect of PPT setting, we establish the first large-scale privacy-preserving portrait matting benchmark named P3M-10k. It contains 10,000 annonymized high-resolution portrait images by face obfuscation along with high-quality ground truth alpha mattes. Specifically, we carefully collect, filter, and annotate about \u003cstrong\u003e10,000\u003c/strong\u003e high-resolution images from the Internet with free use license. There are \u003cstrong\u003e9,421\u003c/strong\u003e images in the training set and \u003cstrong\u003e500\u003c/strong\u003e images in the test set, denoted as \u003cstrong\u003e\u003ci\u003eP3M-500-P\u003c/i\u003e\u003c/strong\u003e. In addition, we also collect and annotate another \u003cstrong\u003e500\u003c/strong\u003e public celebrity images from the Internet without face obfuscation, to evaluate the performance of matting models under the PPT setting on normal portrait images, denoted as \u003cstrong\u003e\u003ci\u003eP3M-500-NP\u003c/i\u003e\u003c/strong\u003e. We show some examples as below, where (a) is from the training set, (b) is from \u003cstrong\u003e\u003ci\u003eP3M-500-P\u003c/i\u003e\u003c/strong\u003e, and (c) is from \u003cstrong\u003e\u003ci\u003eP3M-500-NP\u003c/i\u003e\u003c/strong\u003e.\u003c/p\u003e\n\n\nP3M-10k and the facemask are now \u003cstrong\u003epublished\u003c/strong\u003e!! You can get access to it from the following links, please make sure that you have read and agreed to the agreement. Note that the facemask is not used in our work. So it's optional to download it.\n\n\u003c!-- | Dataset | \u003cp\u003eDataset Link\u003cbr\u003e(Google Drive)\u003c/p\u003e | \u003cp\u003eDataset Link\u003cbr\u003e(Baidu Wangpan 百度网盘)\u003c/p\u003e | Dataset Release Agreement|\n| :----:| :----: | :----: | :----: | \n|\u003cstrong\u003eP3M-10k\u003c/strong\u003e|[Link](https://drive.google.com/uc?export=download\u0026id=1LqUU7BZeiq8I3i5KxApdOJ2haXm-cEv1)|[Link](https://pan.baidu.com/s/1X9OdopT41lK0pKWyj0qSEA) (pw: fgmc)|[Agreement (MIT License)](https://jizhizili.github.io/files/p3m_dataset_agreement/P3M-10k_Dataset_Release_Agreement.pdf)| \n|\u003cstrong\u003eP3M-10k facemask\u003c/strong\u003e (optional)|[Link](https://drive.google.com/file/d/1I-71PbkWcivBv3ly60V0zvtYRd3ddyYs/view?usp=sharing)|[Link](https://pan.baidu.com/s/1D9Kj_OIJbFTsqWfbMPzh_g) (pw: f772)|[Agreement (MIT License)](https://jizhizili.github.io/files/p3m_dataset_agreement/P3M-10k_Dataset_Release_Agreement.pdf)|  --\u003e\n\n| Dataset | \u003cp\u003eDataset Link\u003cbr\u003e(Google Drive)\u003c/p\u003e | \u003cp\u003eDataset Link\u003cbr\u003e(Baidu Wangpan 百度网盘)\u003c/p\u003e | Dataset Release Agreement|\n| :----:| :----: | :----: | :----: | \n|\u003cstrong\u003eP3M-10k\u003c/strong\u003e|[Link](https://drive.google.com/file/d/1odzHp2zbQApLm90HH_Cvr5b5OwJVhEQG/view?usp=sharing)|[Link](https://pan.baidu.com/s/1aEmEXO5BflSp5hiA-erVBA?pwd=cied) (pw: cied) |[Agreement (MIT License)](https://jizhizili.github.io/files/p3m_dataset_agreement/P3M-10k_Dataset_Release_Agreement.pdf)| \n\n\n\n![](demo/p3m_dataset.png)\n\n## P3M-Net and Variants\n\n![](demo/network.png)\n\n\u003cp align=\"justify\"\u003eOur P3M-Net network models the comprehensive interactions between the sharing encoder and two decoders through three carefully designed integration modules, i.e., 1) a tripartite-feature integration (TFI) module to enable the interaction between encoder and two decoders; 2) a deep bipartite-feature integration (dBFI) module to enhance the interaction between the encoder and segmentation decoder; and 3) a shallow bipartitefeature integration (sBFI) module to promote the interaction between the encoder and matting decoder.\u003c/p\u003e\n\n\u003cp align=\"justify\"\u003eWe design three variants of P3M Basic Blocks based on CNN and vision transformers, namely \u003cstrong\u003eP3M-Net(ResNet-34), P3M-Net(Swin-T), P3M-Net(ViTAE-S)\u003c/strong\u003e. We leverage the ability of transformers in modeling long-range dependency to extract more accurate global information and the locality modelling ability to reserve lots of details in the transition areas. The structures are shown in the following figures. \u003c/p\u003e\n\n![](demo/p3m-net-variants.png)\n\n\u003c!-- \u003cp align=\"justify\"\u003eHere we provide the \u003cstrong\u003eP3M-Net(ViTAE-S)\u003c/strong\u003e model pretrained on P3M-10k, and the ViTAE-S and SWIN-T backbones pretrained on ImageNet.\u003c/p\u003e\n\n| Model|  Google Drive | Baidu Wangpan(百度网盘) | \n| :----: | :----:| :----: | \n| P3M-Net(ViTAE-S)  | [Link](https://drive.google.com/file/d/1QbSjPA_Mxs7rITp_a9OJiPeFRDwxemqK/view?usp=sharing) | [Link](https://pan.baidu.com/s/19FuiR1RwamqvxfhdXDL1fg) (pw: hxxy) |\n| ViTAE-S, SWIN-T | [Link](https://drive.google.com/drive/folders/1sBRVgvFNtkZql1_ti_CxCgZs2O8U4RmW?usp=sharing) | [Link](https://pan.baidu.com/s/1pVfUWAs-DcwTbWcUVGrfRw) (pw: wqac) | --\u003e\n\n\n## P3M-CP\n\n\n\u003cp align=\"justify\"\u003eTo further improve the generalization ability of P3M-Net, we devise a simple yet effective Copy and Paste strategy (P3M-CP) that can borrow facial information from publicly available celebrity images without privacy concerns and guide the network to reacquire the face context at both data and feature level, namely P3M-ICP and P3M-FCP. The pipeline is shown in the following figure.\u003c/p\u003e\n\n![](demo/p3m-cp.png)\n\n## Results\n\n\u003cp align=\"justify\"\u003eWe test our network on our proposed P3M-500-P and P3M-500-NP and compare with previous SOTA methods, we list the results as below.\u003c/p\u003e\n\n![](demo/results1.png)\n![](demo/results2.png)\n\n## Quick start - Train\n\nPlease follow this [instruction page](./core/README.md).\n\n## Inference Code - How to Test on Your Images\n\n\u003cp align=\"justify\"\u003eHere we provide the procedure of testing on sample images by our pretrained \u003cstrong\u003eP3M-Net(ViTAE-S)\u003c/strong\u003e model:\u003c/p\u003e\n\n1. Setup environment following this [instruction page](./core/README.md);\n\n2. Insert the path `REPOSITORY_ROOT_PATH` in the file `core/config.py`;\n\n3. Download the pretrained P3M-Net(ViTAE-S) model from here ([Google Drive](https://drive.google.com/file/d/1QbSjPA_Mxs7rITp_a9OJiPeFRDwxemqK/view?usp=sharing) | [Baidu Wangpan](https://pan.baidu.com/s/19FuiR1RwamqvxfhdXDL1fg) (pw: hxxy))) and unzip to the folder `models/pretrained/`;\n\n4. Save your sample images in folder `samples/original/.`;\n    \n5. Setup parameters in the file `scripts/test_samples.sh` and run by:\n\n    `chmod +x scripts/test_samples.sh`\n\n    `scripts/test_samples.sh`;\n\n6. The results of alpha matte and transparent color image will be saved in folder `samples/result_alpha/.` and `samples/result_color/.`.\n\n\u003cp align=\"justify\"\u003eWe show some sample images, the predicted alpha mattes, and their transparent results as below. We use the pretrained \u003cstrong\u003eP3M-Net(ViTAE-S)\u003c/strong\u003e model from section \u003ca href=\"#p3m-net-and-variants\"\u003eP3M-Net and Variants\u003c/a\u003e with `RESIZE` test strategy.\u003c/p\u003e\n\n\u003cimg src=\"samples/original/p_015cd10e.jpg\" width=\"33%\"\u003e\u003cimg src=\"samples/result_alpha/p_015cd10e.png\" width=\"33%\"\u003e\u003cimg src=\"samples/result_color/p_015cd10e.png\" width=\"33%\"\u003e\n\u003cimg src=\"samples/original/p_819ea202.jpg\" width=\"33%\"\u003e\u003cimg src=\"samples/result_alpha/p_819ea202.png\" width=\"33%\"\u003e\u003cimg src=\"samples/result_color/p_819ea202.png\" width=\"33%\"\u003e\n\u003cimg src=\"samples/original/p_0865636e.jpg\" width=\"33%\"\u003e\u003cimg src=\"samples/result_alpha/p_0865636e.png\" width=\"33%\"\u003e\u003cimg src=\"samples/result_color/p_0865636e.png\" width=\"33%\"\u003e\n\n\n## Statement\n\n\u003cp align=\"justify\"\u003eIf you are interested in our work, please consider citing the following:\u003c/p\u003e\n\n```\n@article{rethink_p3m,\n  title={Rethinking Portrait Matting with Pirvacy Preserving},\n  author={Ma, Sihan and Li, Jizhizi and Zhang, Jing and Zhang, He and Tao, Dacheng},\n  journal={International Journal of Computer Vision},\n  publisher={Springer},\n  ISSN={1573-1405},\n  year={2023}\n}\n```\n\n\u003cp align=\"justify\"\u003eThis project is under MIT licence.\u003c/p\u003e\n\nFor further questions, please contact \u003cstrong\u003e\u003ci\u003eSihan Ma\u003c/i\u003e\u003c/strong\u003e at [sima7436@uni.sydney.edu.au](mailto:sima7436@uni.sydney.edu.au) or \u003cstrong\u003e\u003ci\u003eJizhizi Li\u003c/i\u003e\u003c/strong\u003e at [jili8515@uni.sydney.edu.au](mailto:jili8515@uni.sydney.edu.au).\n\n\n## Relevant Projects\n\n\u003ca href=\"https://github.com/ViTAE-Transformer/ViTAE-Transformer-Matting\"\u003e\u003cimg  src=\"https://shields.io/badge/-A_list_of_our_works_in_matting-9cf?style=for-the-badge\"\u003e\u003c/a\u003e\n\u003c/p\u003e\n\n[1] \u003cstrong\u003eDeep Automatic Natural Image Matting, IJCAI, 2021\u003c/strong\u003e | [Paper](https://www.ijcai.org/proceedings/2021/0111.pdf) | [Github](https://github.com/JizhiziLi/AIM)\n\u003cbr\u003e\u003cem\u003e\u0026ensp; \u0026ensp; \u0026ensp;Jizhizi Li, Jing Zhang, and Dacheng Tao\u003c/em\u003e\n\n[2] \u003cstrong\u003ePrivacy-preserving Portrait Matting, ACM MM, 2021\u003c/strong\u003e | [Paper](https://dl.acm.org/doi/10.1145/3474085.3475512) | [Github](https://github.com/JizhiziLi/P3M)\n\u003cbr\u003e\u003cem\u003e\u0026ensp; \u0026ensp; \u0026ensp;Jizhizi Li\u003csup\u003e\u0026#8727;\u003c/sup\u003e, Sihan Ma\u003csup\u003e\u0026#8727;\u003c/sup\u003e, Jing Zhang, Dacheng Tao\u003c/em\u003e\n\n[3] \u003cstrong\u003eBridging Composite and Real: Towards End-to-end Deep Image Matting, IJCV, 2022 \u003c/strong\u003e | [Paper](https://link.springer.com/article/10.1007/s11263-021-01541-0) | [Github](https://github.com/JizhiziLi/GFM)\n\u003cbr\u003e\u003cem\u003e\u0026ensp; \u0026ensp; \u0026ensp;Jizhizi Li\u003csup\u003e\u0026#8727;\u003c/sup\u003e, Jing Zhang\u003csup\u003e\u0026#8727;\u003c/sup\u003e, Stephen J. Maybank, Dacheng Tao\u003c/em\u003e\n\n[4] \u003cstrong\u003eReferring Image Matting, CVPR, 2023\u003c/strong\u003e | [Paper](https://arxiv.org/pdf/2206.05149.pdf) | [Github](https://github.com/JizhiziLi/RIM)\n\u003cbr\u003e\u003cem\u003e\u0026ensp; \u0026ensp; \u0026ensp;Jizhizi Li, Jing Zhang, and Dacheng Tao\u003c/em\u003e\n\n\n[5] \u003cstrong\u003eDeep Image Matting: A Comprehensive Survey, ArXiv, 2023\u003c/strong\u003e | [Paper](https://arxiv.org/abs/2304.04672) | [Github](https://github.com/jizhiziLi/matting-survey)\n\u003cbr\u003e\u003cem\u003e\u0026ensp; \u0026ensp; \u0026ensp;Jizhizi Li, Jing Zhang, and Dacheng Tao\u003c/em\u003e","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fvitae-transformer%2Fp3m-net","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fvitae-transformer%2Fp3m-net","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fvitae-transformer%2Fp3m-net/lists"}