Microsoft Table Transformer HuggingFace Demo

In this video I will explain about Microsoft Table Transformer with a demo.
The Table Transformer model was proposed in PubTables-1M: Towards comprehensive table extraction from unstructured documents by Brandon Smock, Rohith Pesala, Robin Abraham. The authors introduce a new dataset, PubTables-1M, to benchmark progress in table extraction from unstructured documents, as well as table structure recognition and functional analysis. The authors train 2 DETR models, one for table detection and one for table structure recognition, dubbed Table Transformers.
If you like such content please subscribe to the channel here:
www.youtube.co...
If you like to support me financially, It is totally optional and voluntary. Buy me a coffee here: www.buymeacoff...
Relevant Links:
github.com/mic...
huggingface.co...
huggingface.co...
github.com/Nie...
github.com/Nie...

[@senthilchinnappan7631]

For structure recognition, it works on the cropped table, if you feed the entire page, the recognition doesn't produce accurate results.

[@biznessology]

This is really good, I have problem statement, where i am able to detect the table properly but, i dont how to extract the data from there. As i have read the table and save it as csv.

[@anangrajsinghtomar3279]

Thanks for the video. Can you please also provide the resources to fine-tune the "microsoft/table-transformer-detection" model on our custom table dataset? And in what format does this fine-tuning method expect the custom data. I have the custom data in coco format. Does that work ?

[@RitheshSreenivasan]

Please visit the hugging face page or GitHub page and check for the same

[@anangrajsinghtomar3279]

@@RitheshSreenivasan Thanks for replying back. I checked and found out that the most common format is PASCAL VOC XML format. Thanks again for the video.

[@sankettgorey91]

@@anangrajsinghtomar3279 can you help me with the correct dataset format for custom training?

[@iriscodes3983]

Hey @anaagrajsinghtomer hope you find way to fine-tune "Microsoft/table-tranformer-detection" on the custom dataset . Can you please help me too how to fine-tune it, any direct link to colab notebook or way to do it will be appreciated 🙏

[@ujjwalkumarsingh6028]

Hi can you also share the notebook?

[@monkeymaster6489]

How can this be used to convert an image to a CSV?

[@RitheshSreenivasan]

Please have a look at the documentation

[@monkeymaster6489]

@@RitheshSreenivasan Thanks very helpful. I'll probably have a look through the codebase as well. Hey, maybe I should deep dive into the model architecture?

[@monkeymaster6489]

@@RitheshSreenivasan No really, thank you SO MUCH Rithesh

[@RitheshSreenivasan]

@@monkeymaster6489 Yes'

[@adityasinha3851]

how did you convert image to CSV usign this transformer. What techniques did you use ?

[@moibe182]

Great video, an excellent introduction for us who hasn't worked with transformers, thanks! Just one question: at 1:50 you showed the initial documentation in where it can be seen that the result can be a json file, but when you ran the examples it just posted the cell values in the image itself. How can I actually get the json? Thanks!

[@RitheshSreenivasan]

It had been a long time since I made the video, your best bet is their official GitHub

[@t-distributedkid3825]

Great video bro. Liked and subbed. I have been checking this out for last 2 weeks. I have few domain specific problem for which i need few suggestions 1. My tables are spanning 2-3 pages. They are 500X5 sort of tables and doesn't fit fully in a single image. Im stuck here. Any suggestions here would be super helpful

[@RitheshSreenivasan]

Can you do image processing and concatenate the pages into a single large image or try adding headers to every page

[@t-distributedkid3825]

@@RitheshSreenivasan will try the first approach. Have been thinking the same. If it becomes too big won't that become a problem if I want to extract data using ocr or something else

[@RitheshSreenivasan]

@@t-distributedkid3825 Might be. You can also OCR page by page and then apply some rules on text

[@t-distributedkid3825]

@@RitheshSreenivasan ok bro will try thnks

[@tradeNucleus]

Great video.

[@RitheshSreenivasan]

Thanks!

[@thiagarajamuralidaran1371]

Thanks for the video

[@DeepSinghYoutube007]

How can we put a pdf as a input to detect table in a pdf

[@RitheshSreenivasan]

I don’t really remember now. This is a old video. Best to lookup at their GitHub page

[@mahammadfarukhuddin4202]

hi @TanisqStar9830 did you find how to feed pdf file sto extract table data?

[@venkatesanr9455]

Thanks for your videos. If I want to remove the paragraphs & title of the paragraphs by using huggingface models which will be helpful kindly provide some inputs.

[@RitheshSreenivasan]

May be layout LM model can be of use. Let me see if I can make a demo on that

[@venkatesanr9455]

@@RitheshSreenivasan Thanks for your kind replies, sir and kindly do some videos and provide inputs that will be great. I am waiting for your next video, sir