yanghm/convolutional-prototype-learning

An implementation (TensorFlow) of CPL and GCPL appeared in CVPR2018 paper: "Robust Classification with Convolutional Prototype Learning"

convolutional-prototype-learning's Issues

Training Problems

I use the framework in other problem and I find that the data may be close to every class prototypes, do anyone find this problem?

How to choose prototype?

Is there any methodology?

about the datasets

I'd like to ask if anyone can share the author's MNIST. At present, the author hasn't replied to me.Thank you very much!

How can I plot feature representation into 2D spaces? as in figures?

Does paper use t-sne or PCA ?? ..

train model with imagenet dataset

thanks for this great job. your paper just report the minist\cifar scores. do you have train model for imagenet or other largescale dataset? can you share more experience for this largescale image classification task? thank you very much.

Initialization Problem

Thanks for your greate jod. In my experiments, I find it is difficult to converge with prototypes randomly initialized. How can I solve this? I have no prior on the dataset. With a Softmax CNN, the convergence process is stable.

is it right here?

Convolutional-Prototype-Learning/functions.py

Line 14 in 742d60a

dist = f_2 - 2*tf.matmul(features, centers, transpose_b=True) + tf.transpose(c_2, perm=[1,0])