asagar60 / tablenet-pytorch

Pytorch Implementation of TableNet

TableNet-pytorch

Pytorch Implementation of TableNet Research Paper : https://arxiv.org/abs/2001.01469

Description

In this project we will implement an end-to-end Deep learning architecture which will not only localize the Table in an image, but will also generate structure of Table by segmenting columns in that Table. After detecting Table structure from the image, we will use Pytesseract OCR package to read the contents of the Table.

To know more about the approach, refer my medium blog post,

Part 1: https://asagar60.medium.com/tablenet-deep-learning-model-for-end-to-end-table-detection-and-tabular-data-extraction-from-b1547799fe29

Part 2: https://asagar60.medium.com/tablenet-deep-learning-model-for-end-to-end-table-detection-and-tabular-data-extraction-from-a49ac4cbffd4

Data

We will use both Marmot and Marmot Extended dataset for Table Recognition. Marmot dataset contains Table bounding box coordinates and extended version of this dataset contains Column bounding box coordinates.

Marmot Dataset : https://www.icst.pku.edu.cn/cpdp/docs/20190424190300041510.zip Marmot Extended dataset : https://drive.google.com/drive/folders/1QZiv5RKe3xlOBdTzuTVuYRxixemVIODp

Download processed Marmot dataset: https://drive.google.com/file/d/1irIm19B58-o92IbD9b5qd6k3F31pqp1o/view?usp=sharing

Model

We will use DenseNet121 as encoder and build model upon it.

Trainable Params

Download saved model : https://drive.google.com/file/d/1TKALmlwUM_n4gULh6A6Q35VPRUpWDmJZ/view?usp=sharing

Performance compared to other encoder models ( Resnet18, EfficientNet-B0, EfficientNet-B1, VGG19 )

Table Detection - F1

Table Detection - Loss

Column Detection - F1

Column Detection - Loss

Predictions

Predictions from the model

After fixing table mask using contours

After fixing column mask using contours

After processing it through pytesseract

Deployed application

https://vimeo.com/577282006

Future Work

• Deploy this application on a remote server using AWS /StreamLit sharing/heroku.
• Model Quantization for faster inference time.
• Train for more epochs and compare the performances.
• Increase data size by adding data from ICDAR 2013 Table recognition dataset.

References

1. Table Net Research Paper
2. 7 tips for squeezing maximum performance from pytorch
3. StreamLit
4. AppliedAI Course

tablenet-pytorch's Issues

No module named 'model'

I am facing this problem when trying to run on colab. I would be grateful if you could help me how to solve this.

---

ModuleNotFoundError Traceback (most recent call last)

in
13 import pytesseract
14 from io import StringIO
---> 15 from model import TableNet

ModuleNotFoundError: No module named 'model'

---

NOTE: If your import is failing due to a missing package, you can
manually install dependencies using either !pip or !apt.

To view examples of installing some common dependencies, click the
"Open Examples" button below.

Libraries installation and dependencies

Hello, would you be able to provide a mini installation guide of the libraries and their dependencies? Thanks :)

Google drive weights file unavailable

If possible, would you post a new link to the saved model weights? Much appreciated.

No densenet model found

Hi Arun,
I'm not able to find any model within this project, particularly densenet_config_4_model_checkpoint.pth.tar

Could you please provide densenet_config_4_model_checkpoint.pth.tar at least ?

Thanks in advance

I am new in deep learning, can anyone tell me how to finetune the model on my own dataset

Tessar

I found the project very interesting.
but there is an error message.
can you help me.

AttributeError: 'NoneType' object has no attribute 'split'

Unable to Run the project successfully

FileNotFoundError: [Errno 2] No such file or directory: 'densenet_config_4_model_checkpoint.pth.tar'
Traceback:
File "C:\Users\user\Anaconda3\envs\streamlitenv\lib\site-packages\streamlit\scriptrunner\script_runner.py", line 557, in _run_script
exec(code, module.dict)
File "app.py", line 238, in
model = load_model()
File "C:\Users\user\Anaconda3\envs\streamlitenv\lib\site-packages\streamlit\legacy_caching\caching.py", line 573, in wrapped_func
return get_or_create_cached_value()
File "C:\Users\user\Anaconda3\envs\streamlitenv\lib\site-packages\streamlit\legacy_caching\caching.py", line 557, in get_or_create_cached_value
return_value = func(*args, **kwargs)
File "app.py", line 156, in load_model
model.load_state_dict(torch.load("densenet_config_4_model_checkpoint.pth.tar")['state_dict'])
File "C:\Users\user\Anaconda3\envs\streamlitenv\lib\site-packages\torch\serialization.py", line 699, in load
with _open_file_like(f, 'rb') as opened_file:
File "C:\Users\user\Anaconda3\envs\streamlitenv\lib\site-packages\torch\serialization.py", line 230, in _open_file_like
return _open_file(name_or_buffer, mode)
File "C:\Users\user\Anaconda3\envs\streamlitenv\lib\site-packages\torch\serialization.py", line 211, in init
super(_open_file, self).init(open(name, mode))

Model link is dead please share another one

Dependencies

Hello, can you tell me where I can find the dependencies for this project? Searched but did not spot them :) Greetings

could not find MARK

Hi, this project is very helpful for me to write my thesis, can you solve the problem about the installation guide of dependencies for local deployment?

Issue while loading the model

Hi,
While working with the repo,
I faced unpickle error while loading the pre-trained model

.