arXivDaily arXiv每日学术速递 周一至周五更新
北美退伍军人节(Veterans Day),12 日无更新
arXiv 2204.03359cs.CV
收藏

ECCV Caption:通过收集机器与人工验证的图像-标题关联来纠正假阴性,面向 MS-COCO

ECCV Caption: Correcting False Negatives by Collecting Machine-and-Human-verified Image-Caption Associations for MS-COCO

NAVER AI Lab(NAVER AI实验室);Google Research(谷歌研究院);University of Tübingen(图宾根大学)

Sanghyuk Chun, Wonjae Kim, Song Park, Minsuk Chang, Seong Joon Oh

AI总结:

针对图文匹配基准中假阴性缺失关联的问题,构建了机器与人工验证的 ECCV Caption 数据集,提出 mAP@R 排序指标并重评 25 个 VL 模型,发现新指标显著改变了模型排名。

摘要:

图文匹配(ITM)是评估视觉与语言(VL)模型质量的常见任务。然而,现有的 ITM 基准存在一个显著局限:由于数据构建过程本身的原因,它们包含许多缺失的对应关系。例如,一个标题仅与一张图像匹配,尽管该标题也能与其他相似图像匹配,反之亦然。为纠正大量假阴性,我们通过机器和人工标注者补充缺失的关联,构建了 Extended COCO Validation (ECCV) Caption 数据集。我们在标注过程中采用了五种具有不同特性的最先进 ITM 模型。与原始 MS-COCO 相比,我们的数据集提供了 3.6 倍的正向图像到标题关联和 8.5 倍的标题到图像关联。我们还提出使用信息量更大的基于排序的指标 mAP@R,以取代流行的 Recall@K (R@K)。我们在现有和提出的基准上重新评估了现有的 25 个 VL 模型。我们的发现是,现有基准(如 COCO 1K R@K、COCO 5K R@K、CxC R@1)彼此之间高度相关,而当我们转向 ECCV mAP@R 时,排名发生了变化。最后,我们深入探讨了机器标注者选择所带来的偏差影响。源代码和数据集可在 https://github.com/naver-ai/eccv-caption 获取。

英文摘要:

Image-Text matching (ITM) is a common task for evaluating the quality of Vision and Language (VL) models. However, existing ITM benchmarks have a significant limitation. They have many missing correspondences, originating from the data construction process itself. For example, a caption is only matched with one image although the caption can be matched with other similar images and vice versa. To correct the massive false negatives, we construct the Extended COCO Validation (ECCV) Caption dataset by supplying the missing associations with machine and human annotators. We employ five state-of-the-art ITM models with diverse properties for our annotation process. Our dataset provides x3.6 positive image-to-caption associations and x8.5 caption-to-image associations compared to the original MS-COCO. We also propose to use an informative ranking-based metric mAP@R, rather than the popular Recall@K (R@K). We re-evaluate the existing 25 VL models on existing and proposed benchmarks. Our findings are that the existing benchmarks, such as COCO 1K R@K, COCO 5K R@K, CxC R@1 are highly correlated with each other, while the rankings change when we shift to the ECCV mAP@R. Lastly, we delve into the effect of the bias introduced by the choice of machine annotator. Source code and dataset are available at https://github.com/naver-ai/eccv-caption

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。

↑