parseq

parseq copied to clipboard

hello,I want to know how to get 1811 images of the 2077 complete picture in IC2015, because the complete picture downloaded from the official website can see 2077 images, but many models use the 1811 lmdb dataset, and it is impossible to view the specific image used. I tried to remove the non-numeric alphabet pictures from 2077 pictures, and still can't match 1811.So 2077 removes which pictures are confusing.

Hello. We also provide the LMDB for the 1811 subset of IC15. You can just directly extract and save the images from it.

You can check the following references for the filtering done on IC15:

[14] Cheng, Z., Bai, F., Xu, Y., Zheng, G., Pu, S., Zhou, S.: Focusing attention: Towards accurate text recognition in natural images. In: Proceedings of the IEEE international conference on computer vision. pp. 5076–5084 (2017)

[73] Wang, K., Babenko, B., Belongie, S.: End-to-end scene text recognition. In: 2011 International Conference on Computer Vision. pp. 1457–1464. IEEE (2011)

I don't remember the exact details now, but I think IC15 1811 filters for rotated text + non-alphanumeric characters.

Thank you for your reply.