Skip to content
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 our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[CoreML] Solve CoreML frontend issue of image scaler and padding #3800

Merged
merged 1 commit into from Aug 19, 2019

Conversation

@FrozenGene
Copy link
Contributor

commented Aug 18, 2019

We have issue of image scaler and padding in CoreML frontend, so our mobilenet model has incorrect classify result. see issue #3725

Also update the tutorial of CoreML frontend, mobilenet model input image format is BGR.

@FrozenGene FrozenGene force-pushed the FrozenGene:coreml-fix branch from 99f85fb to f5310f3 Aug 18, 2019
@FrozenGene FrozenGene force-pushed the FrozenGene:coreml-fix branch 4 times, most recently from 53ae2ae to 408c314 Aug 18, 2019
…hat Mobilenet mlmodel can work correctly.
@tqchen
tqchen approved these changes Aug 19, 2019
@tqchen tqchen merged commit cb22d50 into dmlc:master Aug 19, 2019
5 checks passed
5 checks passed
continuous-integration/jenkins/pr-merge This commit looks good
Details
windows_mac_build Build #20190818.11 succeeded
Details
windows_mac_build (MacOS_XCode9) MacOS_XCode9 succeeded
Details
windows_mac_build (Windows_VS2017_x64) Windows_VS2017_x64 succeeded
Details
windows_mac_build (Windows_VS2017_x86) Windows_VS2017_x86 succeeded
Details
@tqchen

This comment has been minimized.

Copy link
Member

commented Aug 19, 2019

Thanks @FrozenGene

@FrozenGene

This comment has been minimized.

Copy link
Contributor Author

commented Aug 19, 2019

I find the codebase of master branch's CI is red (http://ci.tvm.ai:8080/blue/organizations/jenkins/tvm/detail/master/1459/pipeline) after merged, however, CI is green as seen when PR is made and I can not reproduce the issue. @tqchen Do we need to restart the CI?

@FrozenGene

This comment has been minimized.

Copy link
Contributor Author

commented Aug 19, 2019

After tried it about ~100 times, I reproduce it. It is one occasionally problem and seems only happened on CUDA (with Cython3 enabled). When we turn on opt_level=3, we call alter_op_layout pass and when we insert one pad with all zeros (i.e. pad(0,0,0,0), we will meet layout problem. However, it shouldn't be. I think I could do one quickfix in coreml frontend. (i.e. When we meet all pad values is zero, we avoid insert pad). I will make one pr tonight. @tqchen

I will try to spare time to construct one reproduce issue and try to find the root cause then fix it.

@FrozenGene FrozenGene deleted the FrozenGene:coreml-fix branch Sep 10, 2019
wweic added a commit to wweic/tvm that referenced this pull request Sep 16, 2019
wweic added a commit to wweic/tvm that referenced this pull request Sep 16, 2019
wweic added a commit to neo-ai/tvm that referenced this pull request Sep 16, 2019
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
2 participants
You can’t perform that action at this time.