New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add Table Transformer #18920
Add Table Transformer #18920
Conversation
The documentation is not available anymore as the PR was closed or merged. |
I'm very much not in favor of adding a new config parameter that controls where the layernorm is applied. I'm not surprised the original code has it, as Facebook AI usually codes models in a modular way, but not Transformers. We had the same thing with BART and friends, and they are coded as distinct models in the library. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Looks good to me! My only concern is that this model might be difficult to publicize as the paper's main contribution is the PubTables-1M dataset and demonstrating how DETR can be used to solve table extraction and related tasks.
Could we add the notebook you are working on to the notebooks repo?
|
||
The Table Transformer model was proposed in [PubTables-1M: Towards comprehensive table extraction from unstructured documents](https://arxiv.org/abs/2110.00061) 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](detr) models, one for table detection and one for table structure recognition, dubbed Table Transformers. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
as well as table structure recognition and functional analysis. The authors train 2 [DETR](detr) models, one for table detection and one for table structure recognition, dubbed Table Transformers. | |
as well as table structure recognition and functional analysis tasks. The authors train two [DETR](detr) models, one for table detection and one for table structure recognition, dubbed Table Transformers. |
|
||
Tips: | ||
|
||
- The authors released 2 models, one for table detection in documents, one for table structure recognition (the task of recognizing the individual rows, columns etc. in a table). |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
- The authors released 2 models, one for table detection in documents, one for table structure recognition (the task of recognizing the individual rows, columns etc. in a table). | |
- The authors released two models, one for table detection in documents, one for table structure recognition (the task of recognizing the individual rows, columns etc. in a table). |
""" | ||
Copy/paste/tweak model's weights to our DETR structure. | ||
""" | ||
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread. Please note that issues that do not follow the contributing guidelines are likely to be ignored. |
Closing this PR in favor of #19614 |
What does this PR do?
This PR adds Table Transformer by Microsoft, which are DETR-compatible models for table detection and table structure recognition tasks in unstructured documents.
Note: I'm making some updates to the original DETR implementation, however these are justified by the fact that the original DETR implementation by Facebook AI also includes these things, which I didn't add when first porting DETR. Hence, our DETR implementation is now more aligned with the original one.
To do: