@cesarpgouveia Thank you for contribution!

Does this problem reproducible with opencv_test_dnn? Or you have run another model?

Hi @alalek !

Does this problem reproducible with opencv_test_dnn? Or you have run another model?

I haven't tested with 'opencv_test_dnn' but I don't think the problem happens in that test. The problem was caused while loading a specific architecture of mine that uses 1x1 convolutions (in order to replace dot layers). However I can try to run opencv_test_dnn in debug tomorrow and see what happens.

Got it.
Looks like this problem is not reproducible with opencv_test_dnn (with used test models) on debug build (unable to reproduce it locally).

I fixed other similar cases too (in more robust way).
Also this patch is rebased on 3.4 branch: https://github.com/opencv/opencv/wiki/Branches
Please check updated patch with your model.

I fixed other similar cases too (in more robust way).

Alright I see what you did on the code changes!

Also this patch is rebased on 3.4 branch: https://github.com/opencv/opencv/wiki/Branches
Please check updated patch with your model.

I didn't fully understand your question. I discover this while testing in the most recent version of opencv (4.5.3). Do you want me to follow this process (rebasing to 3.4, is that it)? https://github.com/opencv/opencv/wiki/Branches#rebasing-a-pull-request-from-master-to-34

Thanks, César.

Do you want me to follow this process (rebasing to 3.4, is that it)?

No. The patch is already rebased to 3.4 branch.

Please check it locally in your environment (you can cherry-pick commit the last commit, no "conflicts" are expected here).