AdvancedLiterateMachinery icon indicating copy to clipboard operation
AdvancedLiterateMachinery copied to clipboard

LORE ONNX Export

Open Holmes2002 opened this issue 11 months ago • 11 comments

Does anyone know how to export LORE ONNX? I applied the same export with CenterNet but without success.

Holmes2002 avatar Mar 03 '24 07:03 Holmes2002

You can try to use the LORE ONNX model in the repo TableStructureRec

SWHL avatar Mar 04 '24 02:03 SWHL

@SWHL Hi, It is great work! However, I just want to export Lore to ONNX and I want to use my own Table Recovery due to low performance when using with large and complex table. Can you share how to convert to ONNX for me ? It's been 3 months and I still haven't converted to ONNX. Hope you can help me!

Holmes2002 avatar Mar 05 '24 09:03 Holmes2002

The LORE onnx model was converted from my previous job, and the conversion envrionment was lost at that time. This weekend, I wiil try to see if I can convert successfully again.

SWHL avatar Mar 06 '24 02:03 SWHL

@Holmes2002 hello, I reorganized the conversion code, and there is no problem in my test here. You can have a try.

This is the repo link: https://github.com/SWHL/ConvertLOREToONNX

SWHL avatar Mar 10 '24 06:03 SWHL

Hi, Thank you for your repo to convert Lore to ONNX. However, seem that you converted model ckpt_wireless-Resnet 18 to ONNX. I want to convert models DLA-34 to ONNX. Have you tried converting the DLA model to ONNX?

Holmes2002 avatar Mar 10 '24 09:03 Holmes2002

I haven't tried DLA before.

SWHL avatar Mar 10 '24 13:03 SWHL

nice

yangy996 avatar Mar 26 '24 08:03 yangy996

@Holmes2002

Have you had any information for DLA-34(onnx)?

phamkhactu avatar Sep 04 '24 01:09 phamkhactu

@phamkhactu No, I successfully converted DLA-34 to ONNX but the inference doesn't work and I don't understand why

Holmes2002 avatar Sep 04 '24 04:09 Holmes2002

@phamkhactu No, I successfully converted DLA-34 to ONNX but the inference doesn't work and I don't understand why

Could you give me snip code for converting onnx if convenience? Or you can compare your onnx with onnx themself. I don't know it extractly which is by a author.

phamkhactu avatar Sep 05 '24 01:09 phamkhactu

@phamkhactu Hi, this is code for converting to ONNX in final file

https://github.com/Holmes2002/Table-Recognition/blob/main/LORE-TSR/src/lib/models/networks/pose_dla_dcn.py

My own inference code but did't work

https://github.com/Holmes2002/Table-Recognition/blob/main/LORE-TSR/src/lib/models/networks/inference_onnx.py

You can check and modify code. Announce to me if you success

Holmes2002 avatar Sep 05 '24 03:09 Holmes2002