TableRecognitionPipelineV2 missed rows in structured table despite correct cell detection

[sagarjgborg]

Hey, I'm using TableRecognitionPipelineV2 in PaddleOCR for table extraction, and although all table cells are detected correctly (bounding boxes are fine), some rows are missing or merged incorrectly in the final structured output (HTML or Excel).

Input table image:

Table cell detection result:

Final excel result

• excel screenshot

grayscale_image_table_1.xlsx

The row POUCH FOLD WIDTH is missed here. Only the first column of that row has been attached to the previous cell.
This happens only for few table is there any way to fix this
I'm using paddleocr 3.1.0

Below is my code

from paddleocr import TableRecognitionPipelineV2

pipeline = TableRecognitionPipelineV2()

output = pipeline.predict('image.png', use_doc_unwarping=False, use_layout_detection=False, use_wired_table_cells_trans_to_html=True)

for res in output:
    res.save_to_img("./output/")
    res.save_to_xlsx("./output/")