arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2103.16002cs.CVcs.CL
收藏

AGQA:一个组合时空推理基准

AGQA: A Benchmark for Compositional Spatio-Temporal Reasoning

University of Pennsylvania(宾夕法尼亚大学);Stanford University(斯坦福大学)

Madeleine Grunde-McLaughlin, Ranjay Krishna, Maneesh Agrawala

摘要:

视觉事件是一个时间动作的组合,涉及空间上与对象交互的演员。在开发能够推理组合时空事件的计算机视觉模型时,我们需要能够分析进展并发现缺点的基准。现有的视频问答基准测试是有用的,但它们往往将多个错误来源合并到一个准确性度量中,并且具有模型可以利用的强烈偏见,因此很难查明模型的弱点。我们提出了行动基因组问答(AGQA),一个新的基准组合时空推理。AGQA包含$9.6K$视频的$192M$不平衡问题答案对。我们还提供了一个平衡的子集的$3.9M$问题答案对,$3$数量级大于现有的基准,最大限度地减少偏见,通过平衡的答案分布和类型的问题结构。虽然人类评估者将我们的问题-答案对中的$86.02\%$标记为正确,但最好的模型只能达到$47.74\%$的准确性。此外,AGQA还引入了多个训练/测试分割来测试各种推理能力,包括对新组合、间接引用和更多组合步骤的概括。使用AGQA,我们评估了现代视觉推理系统,证明了最好的模型几乎没有比利用语言偏见的非视觉基线更好的表现,并且现有的模型都没有推广到训练过程中看不到的新成分。

英文摘要:

Visual events are a composition of temporal actions involving actors spatially interacting with objects. When developing computer vision models that can reason about compositional spatio-temporal events, we need benchmarks that can analyze progress and uncover shortcomings. Existing video question answering benchmarks are useful, but they often conflate multiple sources of error into one accuracy metric and have strong biases that models can exploit, making it difficult to pinpoint model weaknesses. We present Action Genome Question Answering (AGQA), a new benchmark for compositional spatio-temporal reasoning. AGQA contains $192M$ unbalanced question answer pairs for $9.6K$ videos. We also provide a balanced subset of $3.9M$ question answer pairs, $3$ orders of magnitude larger than existing benchmarks, that minimizes bias by balancing the answer distributions and types of question structures. Although human evaluators marked $86.02\%$ of our question-answer pairs as correct, the best model achieves only $47.74\%$ accuracy. In addition, AGQA introduces multiple training/test splits to test for various reasoning abilities, including generalization to novel compositions, to indirect references, and to more compositional steps. Using AGQA, we evaluate modern visual reasoning systems, demonstrating that the best models barely perform better than non-visual baselines exploiting linguistic biases and that none of the existing models generalize to novel compositions unseen during training.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑