RT-DETR-L_wired_cell_det: How to increase max detected objects per image?

[enikeevtg]

Hi PaddleOCR Team!

When using the RT-DETR-L_wired_cell_det model for table cell detection, I encountered a hard limit: the maximum number of objects the model returns per image is 300. This limitation causes problems for images containing large tables, where detecting all cells is critical. In such cases, some cells are simply not detected, which affects the completeness and accuracy of table recognition.

Is there a way to increase the maximum number of detected objects with the current model version (3.1.0)? For example, via a configuration parameter or other method?

I greatly appreciate the PaddleOCR project and the incredible work of its developers.

[Kyo1234567]

Hi guy,
Has your problem about max 300 cells been solved? I, today, also encountered a problem like this. If you have solved this, any help from you will be greatly appreciated!