Wrong bounding box coordinates reported (regression from 4.0.0)

[lorenzob]

• Tesseract Version:

tesseract 4.0.0-279-gec8f
leptonica-1.77.0
libjpeg 6b (libjpeg-turbo 1.2.90) : libpng 1.5.13 : libtiff 4.0.3 : zlib 1.2.7
Found AVX2
Found AVX
Found SSE

• Commit Number:

b67ff53

• Platform:

Linux ip-xxx-xx-xx-xxx.eu-west-1.compute.internal 3.10.0-957.5.1.el7.x86_64 #1 SMP Wed Dec 19 10:46:58 EST 2018 x86_64 x86_64 x86_64 GNU/Linux

Current Behavior:

Space len is reported as -470

Space len is computed as the distance between the right side of previous box and the left side of the one containing the symbol.

Expected Behavior:

Should be about 17 (as in version 4.0.0 and as from pixel inspection)

Test img:

This is a comparison of the app log:

tess-diff.txt

"DEBUG:root:# Spazio troppo breve -470 lo ignoro"

This means "space too short, ignoring it" and the computed distance is -470.

Many other "space len" are reported as negative values.

I think is a new one, not the same as in #2024 and the two referenced issues.

[stweil]

Could you please try to find out when this regression occurred first, maybe by using git bisect? If 4.0.0 was fine, are there other later revisions which still are fine?

[amitdo]

Does this also occur with commit ce88adb?

[lorenzob]

Same problem with ce88adb.

Later I'll try git bisect to find the commit.

[amitdo]

Please try commit 7249571

[lorenzob]

Commit 7249571 works as 4.0.0 (tesseract 4.0.0-29-g7249)

Here is a comparison of the bounding boxes.

4.0.0:

Others:

The boxes visually look fine, even better than the old version. Please let me check my code before spending further time on this, maybe the problem is on my side.

[zdenop]

Did you have change to check your code?

[lorenzob]

Not yet. If you want we can close this and I will reopen it as soon as I have more details.

[zdenop]

With Tesseract-ocr version 4.1.0-rc1-21-g8e83 I got this result:

so I expect problem should be in your code.