TR2026-074

SoREL: Soft-Label Refurbishment with Ensemble Learning for Noisy Long-Tailed Classification Supplementary Material

    •  Hsieh, J.-W., Wu, Y.-H., Hsieh, Y.-K., Li, X., Peng, K.-C., Chang, M.-C., "SoREL: Soft-Label Refurbishment with Ensemble Learning for Noisy Long-Tailed Classification Supplementary Material", CVPR Findings, June 2026.
      BibTeX TR2026-074 PDF
      • @inproceedings{Hsieh2026jun,
      • author = {Hsieh, Jun-Wei and Wu, Ying-Hsuan and Hsieh, Yi-Kuan and Li, Xin and Peng, Kuan-Chuan and Chang, Ming-Ching},
      • title = {{SoREL: Soft-Label Refurbishment with Ensemble Learning for Noisy Long-Tailed Classification Supplementary Material}},
      • booktitle = {CVPR Findings},
      • year = 2026,
      • month = jun,
      • url = {https://www.merl.com/publications/TR2026-074}
      • }
  • MERL Contact:
  • Research Areas:

    Artificial Intelligence, Computer Vision, Machine Learning

Abstract:

This supplementary material provides additional insights and experimental results that complement the main paper. In Section 2, pseudo codes of our Label Refurbishment with Label Raritity Algorithm are described in Algorithm 1. In Section 3, we present additional experimental results comparing our method with Representation CALibration (RCAL). Section 4 explores the noise resistance and bal- ancing capabilities of BANC and Symmetric Cross Entropy (SCE) losses based on the simulated CIFAR-100 dataset. To further clarify our design choices, Section 5 includes an ablation study on the influence of the scaling coefficient cc in Eq. (3) of the main paper. Additionally, 6 examines the impact of the hyperparameter a in Eq. (4) of the main paper. Finally, Section IV 7 details the Rarity Score gi from Eq. (6) of the main paper. These analyses provide a com- prehensive understanding of our proposed framework and its contributions to enhancing the effectiveness of long-tail noisy label learning.

 

  • Related Publication

  •  Hsieh, J.-W., Wu, Y.-H., Hsieh, Y.-K., Li, X., Peng, K.-C., Chang, M.-C., "SoREL: Soft-Label Refurbishment with Ensemble Learning for Noisy Long-Tailed Classification", CVPR Findings, June 2026.
    BibTeX TR2026-075 PDF
    • @inproceedings{Hsieh2026jun2,
    • author = {Hsieh, Jun-Wei and Wu, Ying-Hsuan and Hsieh, Yi-Kuan and Li, Xin and Peng, Kuan-Chuan and Chang, Ming-Ching},
    • title = {{SoREL: Soft-Label Refurbishment with Ensemble Learning for Noisy Long-Tailed Classification}},
    • booktitle = {CVPR Findings},
    • year = 2026,
    • month = jun,
    • url = {https://www.merl.com/publications/TR2026-075}
    • }