Scholarlink

<p>In this paper, we tackle the challenge of real-time target-person matting in multi-person scenes to protect personal privacy. We propose a novel dual encoder-decoder network, DEDNet, which integrates a transformer-based target filter and a CNN-based alpha matte refiner (AMR). The target filter utilizes a reference image to localize the target person and employs a bidirectional cross-attention module and a semi-hidden prior token decoder to enhance target recognition. AMR then utilizes a lightweight CNN to refine the details of the alpha matte. Extensive evaluations on multiple datasets demonstrate that our method outperforms existing state-of-the-art real-time matting approaches while maintaining real-time performance. The source code and pre-trained models are available at <a href="https://github.com/XinJiang1994/TP-Matting">https://github.com/XinJiang1994/TP-Matting</a>, fostering reproducibility and further research in this area.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Privacy-aware Real-Time Target Person Matting in Multi-Person Scenes Using Dual Encoder-Decoder Networks

  • 期刊论文

  • 发表日期：

    2025年6月9日

查看全文
  • Jiang Xin,
  • Xiaonan Fang,
  • Xueling Zhu,
  • Ruyi Dai,
  • Ju Ren,
  • Wenzhen Yue,
  • Yaoxue Zhang

摘要

In this paper, we tackle the challenge of real-time target-person matting in multi-person scenes to protect personal privacy. We propose a novel dual encoder-decoder network, DEDNet, which integrates a transformer-based target filter and a CNN-based alpha matte refiner (AMR). The target filter utilizes a reference image to localize the target person and employs a bidirectional cross-attention module and a semi-hidden prior token decoder to enhance target recognition. AMR then utilizes a lightweight CNN to refine the details of the alpha matte. Extensive evaluations on multiple datasets demonstrate that our method outperforms existing state-of-the-art real-time matting approaches while maintaining real-time performance. The source code and pre-trained models are available at https://github.com/XinJiang1994/TP-Matting, fostering reproducibility and further research in this area.