Scholarlink

SoS (Soccer Segmentation) is a computer vision semantic segmentation dataset for RoboCup soccer. It contains raw images with rectilinear and equisolid projections, metadata for each image with lens and positional information, and coloured segmentation masks. The dataset is generated using Blender and uses 360 \(^\circ \) images from real past RoboCup fields to create a semi-synthetic scene. Our Blender tool used to generate the dataset is available at https://github.com/NUbots/NUpbr . SoS is hosted on HuggingFace and is accessible for download at https://doi.org/10.57967/hf/2099 . We provide benchmarks for SoS using the Visual Mesh and U-Net, and conduct real world experiments on the network with fine-tuning on a small amount of real data.

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

SoS: A Semi-Synthetic RoboCup Soccer Dataset for Visual Segmentation

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Ysobel Sims,
  • Trent Houliston,
  • Matthew Amos,
  • Alexander Biddulph,
  • Jonathan Tabac II,
  • Alana Noonan,
  • Angelique Herfel,
  • Joe Bailey,
  • Johanne Montano,
  • Thomas O’Brien

摘要

SoS (Soccer Segmentation) is a computer vision semantic segmentation dataset for RoboCup soccer. It contains raw images with rectilinear and equisolid projections, metadata for each image with lens and positional information, and coloured segmentation masks. The dataset is generated using Blender and uses 360 \(^\circ \) images from real past RoboCup fields to create a semi-synthetic scene. Our Blender tool used to generate the dataset is available at https://github.com/NUbots/NUpbr . SoS is hosted on HuggingFace and is accessible for download at https://doi.org/10.57967/hf/2099 . We provide benchmarks for SoS using the Visual Mesh and U-Net, and conduct real world experiments on the network with fine-tuning on a small amount of real data.