Merging nearby bounding boxes using Tesseract

wenwenyu / PICK-pytorch

Code for the paper "PICK: Processing Key Information Extraction from Documents using Improved Graph Learning-Convolutional Networks" (ICPR 2020)

https://arxiv.org/abs/2004.07464

MIT License

556 stars 193 forks source link

Merging nearby bounding boxes using Tesseract #8

Closed sophiajwchoi closed 4 years ago

sophiajwchoi commented 4 years ago

Hi Wenwen, In the training data set, I noticed a bounding box includes more than one word. Which program did you use to merge nearby bounding boxes based on distance? Currently, I have a bounding box around each word. Will the accuracy be affected if the bounding boxes are merged together? Thank you!

wenwenyu commented 4 years ago

@sophiajwchoi Bounding boxes are generated by the text detection algorithm. For example, EAST, PSENet, it directly predicts the coordinate of bounding boxes without merging word operation. Tesseract is a general detection method that may not suitable for task-specific detection. I recommend you to use the trained model on your datasets.

tengerye commented 4 years ago

Hi, @wenwenyu, when you refer "text detection", do you mean text line detection? Since text detection has many granularity, from character level to block level. Does your model apply to all levels?