@misc{indiciaee9898bbf20e4,
  title = {MagicSeg: Open-World Segmentation Pretraining via Counterfactural Diffusion-Based Auto-Generation},
  author = {Kaixin Cai and Pengzhen Ren and Jianhua Han and Yi Zhu and Hang Xu and Jianzhuang Liu and Xiaodan Liang},
  year = {2026},
  url = {https://arxiv.org/abs/2603.19575},
  note = {Source identifier: 2603.19575}
}