Segfault with new image and rus+mrz in recent main #4361
Comments
marcreichman-pfi
changed the title
|
Can you run your command under gdb and post stack trace? (debug tess binary) |
|
@egorpugin This is with a debug build, hence the SIGILL instead of SIGSEGV: |
|
Now please download this file |
|
@egorpugin This one works fine now. On a hunch I ran the images from #4148, and those now fail again. I am sure from your comments you understand, the random stuff is masking the various failures that may actually be happening under the hood. |
|
Yes, that is what I'm testing. |
|
It is better to keep that bunch of tickets open until the real issue(s) are fixed. |
|
Try to add a black frame to the image.
|
|
Hi @amitdo, I am aware that subtle changes to the images usually make things work. We've had a series of images while using high throughput through many videos and other images. The segfaults and aborts are popping up from time to time which is why we're filing the tickets. Thanks! |
|
Please try it and tell me if after adding the black frame you still get a segfault. |
|
Hi @amitdo With the border it did not create a segfault. I also tried some others.
All of this makes sense based on things I've seen testing all these tickets. |
|
Like the issues in boxClipToRectangle , Tesseract does not like images where pixels from the foreground touch the image frame. |
|
@marcreichman-pfi, could you please give #4362 (comment) a try? Do you still get crashs with this code change? And similar important: does it change the OCR results? |
|
Hi @stweil - will try it Monday. Thanks! |
|
Hi @stweil, I made the change you suggested locally, and ran through all of my local test files from various tickets. It did not fix the SEGV, but it did fix the abort in #4362. Outputs are slightly modified in most cases but not dramatically so. Full run of tests with Full run of tests with the suggested change: I ran the file from #4362 with my previous workaround for the sigabrt (changing the assert to an if) - this file is Your workaround: My workaround: In the end, I'm not sure which workaround is "better" but they both succeed where the plain Thanks for your continued attention! |
|
definitely the issue should be investigated with this change |
|
Hi @egorpugin - just tried it in this previous comment Thanks! |
|
Yes, I mean we should track down the issue using 0 value instead of any randoms. |
Current Behavior
With the attached image, and recent
main(9f17a3fd) I receive a segfault in Release (SIGILL in Debug) with themrzandruslangages. Both arefast. Rus is official and mrz is available hereRelease:
Debug:
Expected Behavior
No segfault.
Suggested Fix
No response
tesseract -v
Operating System
Ubuntu 22.04 Jammy
Other Operating System
This is running via WSL but it does not appear related to the issue.
uname -a
Linux hostname 5.10.16.3-microsoft-standard-WSL2 #1 SMP Fri Apr 2 22:23:49 UTC 2021 x86_64 x86_64 x86_64 GNU/LinuxCompiler
GCC 11.4
CPU
Intel(R) Core(TM) i7-3720QM CPU @ 2.60GHzVirtualization / Containers
No explicit VM except for WSL2
Other Information
This works in
5.5.0tagThe text was updated successfully, but these errors were encountered: