This algorithm frequently fits character into narrow space ...But I want to add multi lines into broad space about synthtext

[JarveeLee]

Ah ha your solution works pretty nice！Thanks~
After your suggestion averagely it just takes me 6 seconds to select and place text on 1 image with better appearance ! It was bad and took 60 seconds per image yesterday.....

[JarveeLee]

Like this pic,it will place some chinese character around the narrow edge,but I want to add to the broad white space.This situation frequently happens.

[JarveeLee]

sometime like this , I think it is good but this situation seldom happens.

[JarveeLee]

I don't know how to solve this problem....I guess I should change something in text_util.py in line 118
def render_multiline(self,font,text):

But don't know exactly what to do

[JarveeLee]

Maybe it is the problem about depth prediction?

[ankush-me]

@JarveeLee : The segmentation regions are randomly selected for placing text (check here). If you remove this, and instead go through the regions in sequence, the regions should be ordered in decreasing order of their areas -- so then the largest region (as found in the segmentation step), should be selected first for placing text.

Next, regarding rendering multiple lines:
You can change this line, so that text_type is always PARA, which should select multiple lines containing multiple words, based on the dimensions of the segmentation region and the font-size.

[clscy]

Hi, @JarveeLee , I want to generate Chinese scene images through SynthText， I have generated char_freq.cp and font_px2pt.cp based on newsgroup.txt, but it comes somes mistakes.

Traceback (most recent call last):
File "/home/chengls/data/SynthText_06.05/synthgen.py", line 661, in render_text
regions['homography_inv'][ireg])
File "/home/chengls/data/SynthText_06.05/synthgen.py", line 501, in place_text
render_res = self.text_renderer.render_sample(font,collision_mask)
File "/home/chengls/data/SynthText_06.05/text_utils.py", line 366, in render_sample
text = self.text_source.sample(nline,nchar,text_type)
File "/home/chengls/data/SynthText_06.05/text_utils.py", line 606, in sample
return self.fdictkind
File "/home/chengls/data/SynthText_06.05/text_utils.py", line 652, in sample_para
lines = self.get_lines(nline, nword, nchar_max, f=0.35)
File "/home/chengls/data/SynthText_06.05/text_utils.py", line 584, in get_lines
lines = h_lines(niter=100)
File "/home/chengls/data/SynthText_06.05/text_utils.py", line 576, in h_lines
line_start = np.random.choice(len(self.txt)-nline)
File "mtrand.pyx", line 1391, in mtrand.RandomState.choice (numpy/random/mtrand/mtrand.c:14900)
ValueError: a must be greater than 0

So could you tell how to generate Chinese scene images in details? Thank you.

[JarveeLee]

@clscy generate own font_px2pt.cp by adding more chinese fonts which end with .tff,and run invert_font_size.py . And update char_freq.cp based on newsgroup.txt. And you should fix all chinese utf-8 coding problem both in OS and python...

[ankush-me]

@JarveeLee is it possible for you to share your code / data with the changes you made for generating Chinese text? It would be very useful for a lot of users. Thanks!

[JarveeLee]

@ankush-me I can email to you my code next week when go back to company.

[yash0307]

@JarveeLee, would it be possible for you to share code and data with me ? It will be of great help for my ongoing project.

[ankush-me]

@JarveeLee : Is it possible to share it openly, e.g. on GitHub?

[JarveeLee]

@ankush-me I will do this ....Busy with final exam these days...

[JarveeLee]

@ankush-me https://github.com/JarveeLee/SynthText_Chinese_version

[ankush-me]

@JarveeLee Thanks a lot!