Hongbin

Thanks for your interest in our work. The Reverse Order Tracking is designed to utilize only stage two for data association and tracking. This is to avoid redundant bounding boxes...

- The default setting of POINT THRESHOLD = [0, 0, 0] is to avoid computing numbers of points within detection boxes, which would consume a significant amount of time. -...

> > 这个是还没识别完成，你把max_time 调大，最后要出现end{tabular}才算识别完成。表格识别我们是建议有cuda的机器使用的，你有把device设置成cuda吗，识别时长的日志截图看下。正常一个表格用cuda 100秒以内就可以完成 > > 一个表格的提取就这么费吗...一篇文档一般有3-5个，这个解析时间有点恐怖 Sorry for the misunderstanding in the previous description. Based on the issue's results, it should be considered a failure case. However, after...

We have updated StructTable-InternVL2-1B with higher quality HTML and markdown SFT data to enhance the robustness and capabilities of table recognition in both HTML and markdown formats. We welcome you...

Thank you for your interest in our work! Releasing the training code is on our to-do list. We will make it available as soon as possible.

There is no specific timeline. We will complete this work as soon as possible.

Yes. Our model is trained on the DocGenome dataset. Specifically, we extracted the table data from DocGenome to fine-tune our model. Thank you for your interest in our work! Let...

Thank you for your questions. 1. Separate models trained for table and formula recognition. 2. Unlike unimernet, the is no length embedding added to decoder.

We currently utilize the tokenizer from Pix2Struct, but we have expanded the vocabulary to support the Chinese language better.

Thank you for your valuable suggestion. We will continue to improve the model for better performance.