A question about the Table Recognition result reported by paper.
Author: ChunchunwumuCreated Aug 10, 2025Updated Oct 17, 2025
Upon reviewing the details, I noticed that the experiment used 7,904 table images from the PubTabNet test dataset. However, as per public information, the full PubTabNet test dataset contains 9,064 images.
Could you kindly explain the reason for using a subset of the PubTabNet test dataset rather than the entire dataset in your experiment?
Source: bytedance/Dolphin