Network Overfitting and Convert Model

hx173149 / C3D-tensorflow

C3D is a modified version of BVLC tensorflow to support 3D ConvNets.

MIT License

588 stars 262 forks source link

Network Overfitting and Convert Model #5

Closed LongLong-Jing closed 8 years ago

LongLong-Jing commented 8 years ago

Hi. Thanks for your amazing work. This code is a great help for me. I trained this network in UCF101 data set. However I got a training accuracy of nearly 98% and test accuracy of 33%. Do you know what caused the overfitting? Also, can you share how you transfer the caffe model to tensorflow mode? Thanks!

hx173149 commented 8 years ago

yes if you want to get a high accuracy, you can look at this issue： https://www.dropbox.com/sh/8wcjrcadx4r31ux/AAAkz3dQ706pPO8ZavrztRCca?dl=0

LongLong-Jing commented 8 years ago

Thanks. I will download this model and train again. Also can you give me a hint about how to convert caffe model to tensorflow model? I have struggling about converting model for a long time. Thanks!

hx173149 commented 8 years ago

@LongLong-Jing read the caffe model to arrays, and saved it to npy format files, reload these npy files to tensorflow

LongLong-Jing commented 8 years ago

@hx173149 Thanks!