VCR-Bench：视频分类鲁棒性模块化开源基准
原标题：VCR-Bench: A Modular Open-Source Benchmark for Video Classification Robustness
AI 摘要
研究团队发布 VCR-Bench，一个面向视频分类鲁棒性的模块化开源基准框架，旨在填补图像分类有成熟对抗鲁棒性评测而视频分类缺乏统一基础设施的空白。该框架统一了视频加载、分类器封装、对抗攻击与防御、感知指标、配置预设和结果日志，目前集成 30 个视频分类模型、14 种对抗攻击和 10 种防御方法，并在 Kinetics-400 子集上报告干净准确率、攻击成功率、感知质量、运行时间和内存占用。代码与可复现脚本已在 GitHub 开源。
正文节选
VCR-Bench: A Modular Open-Source Benchmark for Video Classification Robustness Abstract. Robustness of image classification has several benchmarks, but their video counterparts are absent. In video classification temporal dimension introduces additional degrees of freedom for adversarial attacks, defenses, and preprocessing. Temporal sampling, perturbation budgets, and metric aggregation also interact in ways with no direct analogue in the image setting. Therefore, robustness for video classifie