Open Anang1502 opened 10 months ago
Hi,
Is this with one of the pre-trained models and one of the released datasets? If so can you share which model and which input image?
If not, is there any other information you can share so we can understand what caused this bug?
Best, Brandon
did you solve this problem,bro?
Hi @bsmock, Sorry for the late response on this. Yes, this issue occurred in the pre-trained model TATR-v1.1-Pub provided in the official repo. I was testing it on my data. This issue occurs when the pre-trained structure model gives y0 coordinate of the predicted table row greater than y1 and the code fails while running the assert statement for the same check.
Hi @Shaobo-Zhao, I didn't find any permanent solution for this but as a temporary fix, I assigned y0 = y1 wherever I encountered y0 > y1. This issue happens rarely for 1 or 2 images randomly and I could not find any specific pattern for this. But since the training fails even if it occurs for 1 image so I have applied this temporary fix for now.
thank you ,my bro. i will try as you said.
best wishes to you.
hi bro ,
guess what
installed pillow==8.3.1
the problem was solved.
nice
For one of the image, the table transformer structure model gives y0>y1 for row coordinates. This causes issues in further processing the table. What might be causing this issue?