Issue with using different model

Question

Issue with using different model

mpbharat opened this issue 8 years ago · comments

Hi
I'm trying use converted caffe models and other mxnet models in place of Inception-BN Network model. The app crashes at execution of MXPredForward(predictor); during p->exec->Forward(false); . Anyone else faced the same issue? Any help would be appreciated.

Thanks for the amazing app

Michele Toni · Answer 1 · Fri Sep 30 2016 15:46:33 GMT+0800 (China Standard Time)

I can't be sure if it works for you, but I suggest you to update the core mxnet. If you use the original mxnet_predict-all.cc you have to create a new one following this guide https://github.com/dmlc/mxnet/tree/master/amalgamation . I had to do so to use newer model architectures.

Yuan Zhang · Answer 2 · Thu Jun 08 2017 23:02:45 GMT+0800 (China Standard Time)

Hi, @SlipknotTN
I faced the similar problem, I guess the problem was caused by the out-of-date mxnet_predict-all.cc. I wanna know, after updating mxnet_predict-all.cc, did it work? Can you share your mxnet_predict-all.cc?

Michele Toni · Answer 3 · Mon Jun 19 2017 16:27:12 GMT+0800 (China Standard Time)

This is my latest tested version, but it is 1 year's old, so I suggest you to run the amalgamation from scratch. Follow the iOS instructions here https://github.com/dmlc/mxnet/tree/master/amalgamation . Updating mxnet_predict-all I resolved the problem and I managed to run a model not supported by the version present in this repo. I tested a SqueezeNet network.

mxnet_predict-all.cc.zip

Yuan Zhang · Answer 4 · Mon Jun 19 2017 23:26:33 GMT+0800 (China Standard Time)

Thanks for your help! In fact, I follow the instructions you mentioned and create the prediction file, and it works on iOS. If you need the file, I can share with you.

Haoxiang Li · Answer 5 · Fri Jun 23 2017 13:34:27 GMT+0800 (China Standard Time)

Thanks guys. Sorry that I don't have time to maintain the mxnet file.