Skip to content

Latest commit

 

History

History
93 lines (79 loc) · 5.3 KB

README.md

File metadata and controls

93 lines (79 loc) · 5.3 KB
Raw

Two-Stage Cuneiform Sign Detection (Detecting Bounding Boxes + Image Classification)

Data+Code is part of Paper Sign Detection for Cuneiform Tablets from Yunus Cobanoglu, Luis Sáenz, Ilya Khait, Enrique Jiménez.
Data on Zenodoo DOI. See https://github.com/ElectronicBabylonianLiterature/cuneiform-ocr/blob/main/README.md for overview and general information of all repositories associated with the paper from above (inlcuding one-stage model).

This Repository based on mmocr and mmcv can be used for training the Model once trained one has to use a different repository (https://github.com/ElectronicBabylonianLiterature/cuneiform-ocr) based on mmdetection to unify both models to get the final model. This is due to dependencies of the two repositories. For the data use https://github.com/ElectronicBabylonianLiterature/cuneiform-ocr-data.

Checkpoints and data can be found here DOI. One needs to download ready-for-training.tar.gz and efficient-net and fcenet-net modelconfig and checkpoints. Ready-for-training contains several folders which can be copy pasted according to instructions below. For training the models one can download FCENet Weights pretrained on Icdar2015 from mmocr documentation and EfficientNet Weights pretrained on ImageNet from mmpretrain documentation.

The checkpoints of the pretrained models can be found here: https://syncandshare.lrz.de/getlink/fi39rfQ11LtbxEBTmwVs2u/ There should be folder cuneiform-ocr-classification-detection/checkpoints/ with all checkpoints

Requirements (Tested with Python 3.11 )

  • torch==2.0.1, torchvision==0.15.2 (pip install torch==2.0.1 torchvision==0.15.2 torchaudio==2.0.2) https://pytorch.org/get-started/previous-versions/
  • pip install -U openmim
  • mim install mmengine (tested with 0.8.3)
  • mim install "mmcls==1.0.0rc5" (installing as a dependency "mmcv==2.0.0")
  • mim install "mmdet==3.0.0rc6"
  • mim install mmocr #tested with version 1.0.1
  • pip install -r requirements.txt

Detection

Training

  • cd cuneiform_ocr/detection

  • have data in cuneiform_ocr/detection/data

  • data

    • icdar2015
      • textdet_imgs
      • textdet_test.json
      • textdet_train.json

  • python3 mmocr_tools/train.py custom_configs/fcenet_dcvn_debug.py #validation set is used after one epoch to make sure everything is working

  • python3 mmocr_tools/train.py custom_configs/fcenet_dcvn.py #validation set is used after 50 epochs

Testing

  • cd cuneiform_ocr/detection
  • python3 mmocr_tools/test.py custom_configs/fcenet_dcvn.py ../../checkpoints/fcenet_resnet50-dcnv2.pth (replace checkpoints with trained checkpoints)

Classification

Training

  • cd cuneiform_ocr/classification
  • have data in cuneiform_ocr/classification/data
  • data
    • ebl
      • test_set
      • train_set
      • test.txt
      • train.txt
      • classes.txt (optional)
  • python3 mmclassification_tools/train.py custom_configs/efficient_net.py

Testing

  • python3 mmclassification_tools/test.py custom_configs/efficient_net.py ../../checkpoints/efficientnet-b0_3rdparty_8xb32-aa-advprop_in1k_20220119-26434485.pth

Errors

Sometimes trying to install mim install "mmcv==2.0.0rc4" can take forever and it fails if the pytoch version and cuda mismatch. In this case skip installation of mmcv and install mim install "mmcls==1.0.0rc5" which as a dependency has mmcv and will install it for you. (You can try using torch cpu installing everything and then uninstall torch cpu and install torch with cuda to dotch the pytorch version and cuda mismatch error). Now when running a script you may get this error:

Traceback (most recent call last):
  File "/home/yunus/PycharmProjects/cuneiform-ocr-classification-detection/cuneiform_ocr/classification/mmclassification_tools/train.py", line 12, in <module>
    from mmcls.utils import register_all_modules
  File "/home/yunus/PycharmProjects/cuneiform-ocr-classification-detection/.venv/lib/python3.11/site-packages/mmcls/__init__.py", line 18, in <module>
    and mmcv_version < digit_version(mmcv_maximum_version)), \
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

In which case just comment the assert statement in .venv/lib/python3.11/site-packages/mmcls/__init__.py and everything should work.

img shape error needs in results['img_shape'] = img.shape[:2] replace line 102 img = mmcv.imfrombytes( img_bytes, flag=self.color_type, backend=self.imdecode_backend) with img = mmcv.imfrombytes( img_bytes, flag=self.color_type, backend="pillow") of file .venv/lib/python3.11/site-packages/mmcv/transforms/loading.py in mmcv

Cite this paper

@article{CobanogluSáenzKhaitJiménez+2024,
url = {https://doi.org/10.1515/itit-2024-0028},
title = {Sign detection for cuneiform tablets},
title = {},
author = {Yunus Cobanoglu and Luis Sáenz and Ilya Khait and Enrique Jiménez},
journal = {it - Information Technology},
doi = {doi:10.1515/itit-2024-0028},
year = {2024},
lastchecked = {2024-06-01}
}