重新审视持续视觉问答：面向多模态持续学习的公平与鲁棒评测
Re-evaluating Continual VQA: Toward Fair and Robust Evaluation for Multimodal Continual Learning
对做多模态持续学习的人，这篇论文指出共享答案词表与静态答案先验会系统性高估性能、低估遗忘，并给出可复用的无偏基准构造方式（token 级不相交答案空间 + 任务内训练-测试分布偏移）；其 question-only replay 与双层蒸馏基线也可作为轻量防遗忘的对照方案。
原文依据 shared answer vocabularies across tasks encourage answer memorization, inflating performance and underestimating forgetting
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论持续视觉问答（Continual VQA）为多模态持续学习带来了独特挑战，要求模型在增量获取新知识的同时，保持跨任务的视觉-语义定位。然而，现有基准阻碍了对这类能力的公平与鲁棒评测，因为它们允许模型利用数据集偏差，而非展现真正的持续推理。我们指出当前基准设计中的两个结构性缺陷。第一，任务间共享的答案词表鼓励答案记忆，从而虚高性能并低估遗忘。第二，每个任务内静态的答案先验使训练与测试的答案分布几乎相同，掩盖了分布偏移下的鲁棒性。为解决这些问题，我们提出 UCo-VQA，一个无偏基准套件，它在任务间强制实施 token 级不相交的答案空间，并引入任务内训练-测试分布偏移，从而能够更公平地评估多模态持续学习中的遗忘与泛化。我们进一步提供一个参数高效基线，通过仅问题回放与双层蒸馏来缓解遗忘并增强视觉定位，为持续适应提供一个轻量且内存高效的框架。在 UCo-VQA 上的大量实验表明，先前方法在有偏设定下大幅高估了性能，而我们的方法取得了最优结果，鲁棒性与保持率分别最高提升 4.18% 与 2.21%。
英文原文摘要 1536 字符
Continual Visual Question Answering (Continual VQA) poses unique challenges for multimodal continual learning, requiring models to incrementally acquire new knowledge while preserving visual-semantic grounding across tasks. However, existing benchmarks hinder fair and robust evaluation of such capabilities, as they allow models to exploit dataset biases rather than demonstrate genuine continual reasoning. We identify two structural flaws in current benchmark design. First, shared answer vocabularies across tasks encourage answer memorization, inflating performance and underestimating forgetting. Second, static answer priors within each task make the training and test answer distributions nearly identical, obscuring robustness under distribution shifts. To address these issues, we introduce UCo-VQA, an Unbiased benchmark suite that enforces token-level disjoint answer spaces across tasks and introduces intra-task train-test distribution shifts, enabling fairer assessment of forgetting and generalization in multimodal continual learning. We further provide a parameter-efficient baseline that mitigates forgetting and enhances grounding through question-only replay and dual-level distillation, offering a lightweight and memory-efficient framework for continual adaptation. Extensive experiments on UCo-VQA reveal that prior methods substantially overestimate performance under biased setups, while our approach achieves state-of-the-art results, improving robustness and retention by up to 4.18% and 2.21%, respectively.