search

DevashishPrasad / CascadeTabNet

This repository contains the code and implementation details of the CascadeTabNet paper "CascadeTabNet: An approach for end to end table detection and structure recognition from image-based documents"
MIT License
1.47k stars 422 forks source link

Table structure recognition is not predicted for second table of demo image #44

Closed nazarimilad closed 4 years ago

nazarimilad commented 4 years ago

First and foremost, thanks for this interesting paper and also this repository!

Now, as you can see in the README, in the demo gif not only both tables are detected but structure recognition is successful for both tables (in the last step of the animation).

However, when predicting this demo image, I get the different results:

image

As you can see in the screenshot, both tables are detected succesfully. But in the right table no cell is recognised. In the left table, cells in the last columns are also not recognised. I'm using the same checkpoint file and configuration as in the demo Jupyter notebook. I tried lowering the threshold, but that didn't help.

How can I improve the prediction so that I get the same performance as shown in the demo gif? Am I missing some postprocessing, or am I not using the optimal configuration, or something else? I'm not sure, I hope you could help.

Thanks!

DevashishPrasad commented 4 years ago

Hello @nazarimilad

This is the perfect output from the model. After this you need to do the post-processing. For details about the postprocessing please refer the CascadeTabNet/Table Structure Recognition/README.md and find the code for the same in CascadeTabNet/Table Structure Recognition/.

Thanks