contribute into gluoncv

xilaili / AOGNet

Code for CVPR 2019 paper: " Learning Deep Compositional Grammatical Architectures for Visual Recognition"

MIT License

129 stars 20 forks source link

contribute into gluoncv #1

Open mli opened 5 years ago

mli commented 5 years ago

Very interesting work! I’m wonder if you are interested in reimplement the symbol codes in gluon and contribute into gluoncv https://gluon-cv.mxnet.io ?

xilaili commented 5 years ago

Thanks. We tried the gluon before, but we got a problem that the training is not optimizable with our big models (small model works fine). I guess the problem is the initialization. We're planning to reimplement it with your latest changes and see if the problem can get fixed. I'll let you know when it's done. If we still got the same problem, I'll ask you for help.

mli commented 5 years ago

Sounds great!