tf and theano backends give different results for pretrained models.

coxlab / prednet

Code and models accompanying "Deep Predictive Coding Networks for Video Prediction and Unsupervised Learning"

https://arxiv.org/abs/1605.08104

MIT License

759 stars 259 forks source link

Closed zym1010 closed 5 years ago

zym1010 commented 5 years ago

Hi,

seems that using tensorflow backend gives worse results than using tf backend, for the provided pretrained models (which are trained on theano).

I think this is because when loading weights, conv kernels are not flipped as they should be.

bill-lotter commented 5 years ago

Thanks! I added tensorflow and theano weights now to the model download