DISPBENCH
收藏资源简介:
DISPBENCH是一个全面评估视差估计方法可靠性的基准测试工具。它通过评估方法在对抗攻击和由2D Common Corruptions引起的分布偏移方面的鲁棒性,系统地评估了多种数据集和不同的破坏场景。DISPBENCH支持多个数据集,包括FlyingThings3D和KITTI2015,并允许使用5种不同的对抗攻击和15种常见的破坏进行评估。DISPBENCH旨在解决当前视差估计方法在实际应用中可靠性不足的问题,通过提供一套评估工具,帮助研究人员构建更鲁棒、更通用的模型。
DISPBENCH is a benchmark tool for comprehensively evaluating the reliability of disparity estimation methods. It systematically assesses multiple datasets and diverse corruption scenarios by measuring the robustness of these methods against adversarial attacks and distribution shifts induced by 2D Common Corruptions. DISPBENCH supports a range of datasets including FlyingThings3D and KITTI2015, and allows evaluation with 5 distinct adversarial attacks and 15 common corruptions. DISPBENCH aims to address the insufficient reliability of current disparity estimation methods in practical applications, and provides a comprehensive evaluation toolkit to help researchers develop more robust and generalizable models.

- 1DispBench: Benchmarking Disparity Estimation to Synthetic Corruptions德国曼海姆大学数据与网络科学小组, 德国萨尔州信息学园区马克斯·普朗克计算机科学研究所 · 2025年



