NotificationsYou must be signed in to change notification settings
Fork56.4k
Star85.3k

DNN : fix bug in extracting prior-box variances in detection output layer#20525

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to ourterms of service andprivacy statement. We’ll occasionally send you account related emails.

Already on GitHub?Sign in to your account

Jump to bottom

Merged

opencv-pushbot merged 1 commit intoopencv:3.4fromjebastin-nadar:fix-prior-variances

Aug 13, 2021

Merged

DNN : fix bug in extracting prior-box variances in detection output layer#20525

opencv-pushbot merged 1 commit intoopencv:3.4fromjebastin-nadar:fix-prior-variances

Aug 13, 2021

Conversation

Copy link

Contributor

jebastin-nadar commentedAug 9, 2021

GetPriorBBoxes() assumes the input blob has shape (1, 2, num_priors * 4, 1) with the second channel being prior-box variances. But when_varianceEncodedInTarget is true, the input has shape (1, 1, num_prioirs * 4, 1), as checked by the following assertion

opencv/modules/dnn/src/layers/detection_output_layer.cpp

Line 239 in424eaba

CV_Assert(inputs[2][1] ==1 + (int)(!_varianceEncodedInTarget));

Another reference in thereshape for the input of Detection output layer in faster-rcnn-vgg16.

Pull Request Readiness Checklist

See details athttps://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request

I agree to contribute to the project under Apache 2 License.
To the best of my knowledge, the proposed patch is not based on a code under GPL or other license that is incompatible with OpenCV
The PR is proposed to proper branch
There is reference to original bug report and related work
There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
The feature is well documented and sample code can be built with the project CMake

fix bug in prior-box variances

2a17705

asmorkalov added the category: dnn label

Aug 9, 2021

asmorkalov requested a review fromJulieBar

August 9, 2021 07:13

Copy link

Member

alalek commentedAug 9, 2021

@SamFC10 Thank you for contribution!
Is it possible to add accuracy test for this case?

Copy link

ContributorAuthor

jebastin-nadar commentedAug 9, 2021

Is it possible to add accuracy test for this case?

There were some faster rcnn tests which crashed sporadically in my int8 PR (#20228) on Windows CI builders and local windows machine with the message -unknown file: error: SEH exception with code 0xc0000005 thrown in the test body. Tracing back, it was the illegal memory access in prior-box variances which caused these crashes. I don't know how to reproduce these sporadic crashes in a test. Will need some help.

asmorkalov removed the request for review fromJulieBar

August 11, 2021 06:50

Copy link

Member

alalek commentedAug 11, 2021

Tracing back, it was the illegal memory access in prior-box variances which caused these crashes.

This usually happens when memory is accessed without checking arrays layouts (through CV_Check* or CV_Assert macros). It makes sense to add these checks there.