flip_filters = True / False

Ignotus / theano-flownet

FlowNetS and FlowNetC port to Theano

MIT License

8 stars 4 forks source link

flip_filters = True / False #2

Closed htzheng closed 6 years ago

htzheng commented 7 years ago

When training the theano-flownet, I observed that using the original code causes exploding gradients. However, after setting _flipfilters option in leaky_deconv() and upsample() layer to False, the Theano training works.

So I just wonder is the _flipfilters=true necessary for leaky_deconv() and upsample() ? Is this option for importing Caffe weights?

Thank you

Ignotus commented 7 years ago

Hi,

I am not sure about that, if you do with flip_filters=False a network will produce quite poor results on optical flow.