Detect mutli-line text in one box? about pixel_link HOT 8 OPEN

你好，如何识别里面的文字啊？

from pixel_link.

@Heterfire 可以使用crnn

from pixel_link.

If your data is horizontally aligned, try adding weights to your link_conf_thresholds. Increase vertical conf_thesh and decrease horizontal conf_thresh.
Works like a charm.

from pixel_link.

@aravinthmuthu could you kindly share specifics on how you were able to add weights to your link_conf_thresholds?

from pixel_link.

I'd like to know how to increase vertical conf_thesh and decrease horizontal conf_thresh as well. I looked into the code, but I could not find it. Anyone can help?
Thanks.

from pixel_link.

I found that we can set each neighbour pixels threshold like:
link_conf_threshold = np.array([0.8, 0.8, 0.8, 0.45, 0.45, 0.8, 0.8, 0.8])
But I don't think we can set the link_conf_threshold vertically and horizontally.

from pixel_link.

I found that we can set each neighbour pixels threshold like:
link_conf_threshold = np.array([0.8, 0.8, 0.8, 0.45, 0.45, 0.8, 0.8, 0.8])
But I don't think we can set the link_conf_threshold vertically and horizontally.

Hello, may I ask where you changed the link_conf_threshold to an array, is it supposed to be a numeric value?

from pixel_link.

@dajiangxiaoyan how did you re-train your model inorder to detect multi-line in one box. What are the changes needed to be made to achieve that?

from pixel_link.