Table detection and table structure recognition using Yolov5/Yolov8, and you can get the same (even better) result compared with Table Transformer (TATR) with smaller models.
You can download PubTables-1M from Microsoft Research Open Data, and uncompress PubTables-1M-Image_Page_Detection_PASCAL_VOC.tar.gz and PubTables-1M-Image_Table_Structure_PASCAL_VOC.tar.gz to the directory data/pubtables-1m/. Or you can download PubTables-1M, FinTabNet.c and ICDAR-2013.c using data/download_data.sh.
Then, you can run the 4 notebooks (data/voc2coco_detection.ipynb, data/voc2coco_structure.ipynb, data/voc2coco_structure_fintabnet.ipynb, and data/voc2coco_structure_icdar2013.ipynb) to convert VOC format to COCO format.
You can clone latest Yolov5 code from https://github.com/ultralytics/yolov5 to the directory yolov5/, and run the 2 scripts to train table detection model (yolov5/train_PubTables-1M_detection.sh) and table structure recognition model (yolov5/train_PubTables-1M_structure.sh). You may need to change the variable path of the yaml files according to your environment in the directory yolov5/data/.
I have trained each model using yolov5s for 10 epochs, and you can use the models in the directory yolov5/runs/ for fast try or finetune from the checkpoints.
You can change to the directory yolov8/, and run the 2 scripts to train table detection model (yolov8/train_PubTables-1M_detection.sh) and table structure recognition model (yolov8/train_PubTables-1M_structure.sh). You may need to change the variable path of the yaml files according to your environment in the directory yolov8/data/.
I have trained each model using yolov8s for 10 epochs, and you can use the models in the directory yolov8/runs/detect/ for fast try or finetune from the checkpoints.
You can run the notebook table_structure_recognition.ipynb to convert a table image to an excel file. Please pay attention to the ocr function, you should use all-in-one-ai, or PaddleOCR, or any OCR service to get the ocr result.