… the object bias in imagecaptioning models. To reduce objecthallucination in imagecaptioning, … to be treated as ground truth to train imagecaptioning models. By extensive analysis, we …
M Feng, Y Tang, Z Zhang, C Xu - arXiv preprint arXiv:2406.12663, 2024 - arxiv.org
… LVLMs often suffer from a significant challenge known as hallucination (Rohrbach et al.… actual objects present in the input image. In this work, we mainly consider objecthallucination. …
H Ben, S Wang, M Wang, R Hong - Proceedings of the 2024 …, 2024 - dl.acm.org
… For iteration refinement, we hallucinate the generated captions with objects to construct the pseudo-matched sentences to refine the generator, which incrementally increases the …
N Ullah, PP Mohanta - Proceedings of the Asian …, 2022 - openaccess.thecvf.com
… is known as a hallucination [27,38,24] in the literature. Unlike imagecaptioning, there are two types of hallucination in the case of video captioning: objecthallucination occurs when the …
… Work in this paper indicates that features based on object detection model output and imagecaptioning model output may be a good starting point for designing such a more …
L Wang, J He, S Li, N Liu, EP Lim - International Conference on Multimedia …, 2024 - Springer
… that enables instruction-tuned LVLMs to reduce fine-grained objecthallucination by fine-tuning them on additional rewritten captions derived from curated high-quality imagecaptions. …
… how human captioning attention can strengthen imagecaptioning. … Recent progress on imagecaptioning is driven by the development of … Objecthallucination in imagecaptioning …
… that imagecaptioning models sometimes hallucinate words … bias imagecaptioning models from frequently occurring object … generating frequent object words rather than hallucinate them…
… We address the challenges faced in this ield by emphasizing issues such as object hallucination, missing context, illumination conditions, contextual understanding, and referring …