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

[FRONTEND][MXNET] Fix L2normalization error when run ssd model #1527

Closed
wants to merge 949 commits into from

Conversation

ndcuong91
Copy link

This PR fix error L2Normalization when run ssd models from MXNET frontend. I also faced another error when run it with target openCL as described here https://github.com/dmlc/tvm/issues/1302. For short term solution, i reduced num_thread_x = 8 (line 307) and ofactor = 4 (line 309) in topi/python/topi/cuda/conv2d_nchw.py and i could run my ssd model successfully. How about long term solution for this issue?

eric-haibin-lin and others added 30 commits June 10, 2018 20:49
* Fix a link in android_deploy/README.md and a error while building android_deploy.

* revert and change APP_STL in Application.mk
* add support for subgraphs.

* fix.

* fix.

* Fix compilation error

* Fix compilation error

* add comments.

* update comments.

* Sanity check on subgraphs when creating IndexedGraph

* avoid the overhead of sanity check.

* Stop using non-recursive DFS

* Trigger CI

* trigger CI
@kaishijeng
Copy link

@kevinthesun

@kevinthesun
Copy link
Contributor

kevinthesun commented Aug 2, 2018

@kaishijeng Yes. L2Normlization is required for gluoncv ssd implementation: https://github.com/dmlc/gluon-cv/blob/master/gluoncv/model_zoo/ssd/ssd.py I plan to support deploying these models in the next few weeks.

@kaishijeng
Copy link

@kevinthesun

Do you plan to support OpenCL for mobilenet ssd?

Thanks,

@yzhliu
Copy link
Member

yzhliu commented Aug 6, 2018

Please rebase according to #1394
Sorry for the inconvience

@kevinthesun
Copy link
Contributor

@kaishijeng I think @Laurawly is working for OpenCL SSD support.

@yzhliu yzhliu added the status: need update need update based on feedbacks label Aug 13, 2018
@yzhliu
Copy link
Member

yzhliu commented Aug 13, 2018

@titikid Could you update the status? wondering whether @Laurawly 's recent work fixes problems #1578

@ndcuong91 ndcuong91 requested a review from yzhliu as a code owner August 15, 2018 04:45
@ndcuong91
Copy link
Author

@yzhliu i updated l2_normalize's implementation with test cases. Plz help me review it. Thanks!

@yzhliu
Copy link
Member

yzhliu commented Aug 16, 2018

@titikid need to rebase according to #1394 . currently the PR has too many unrelated stuffs to review. Sorry for the inconvenience.

@ndcuong91
Copy link
Author

@yzhliu sorry. I got some issues when rebasing. May i make another PR?

@ndcuong91 ndcuong91 closed this Aug 20, 2018
@kaishijeng
Copy link

@kevinthesun

Any progress of gluoncv ssd implementation: ?

Thanks,

@kevinthesun
Copy link
Contributor

gluoncv ssd will be supported through hybrid frontend. It is delayed due to some limitation of hybrid frontend.

@madhavajay
Copy link

@kevinthesun Anything that we can help with?

@kaishijeng
Copy link

@kevinthesun

Any progress on this?

Thanks,

@kaishijeng
Copy link

@titikid @kevinthesun

Does current TVM code support mobilenet-ssd yet?

Thanks,

@kevinthesun
Copy link
Contributor

kevinthesun commented Dec 2, 2018

Have you tried mxnet_ssd model? I think that should be supported now. Gluoncv models are in progress.

@kaishijeng
Copy link

Does it have mobilenet_ssd?
Thanks,

@kevinthesun
Copy link
Contributor

https://github.com/zhreshold/mxnet-ssd contains mobilenet ssd.

@kaishijeng
Copy link

I am not able to run deploy_ssd.py successfully. So I don't think I can try mobilenet_ssd using deploy_ssd.py

Thanks,

@kevinthesun
Copy link
Contributor

You can file a issue on discuss forum.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
status: need update need update based on feedbacks
Projects
None yet
Development

Successfully merging this pull request may close these issues.