Good NEWS ! We have released the all the codes. Please refer to the project Hard-Aware-Deeply-Cascaded-Embedding_release

Yuan Y, Yang K, Zhang C. Hard-Aware Deeply Cascaded Embedding[J]. arXiv preprint arXiv:1611.05720, 2016.

This is the raw code for our work submitted to cvpr-2017. we will release the complete version in the future.(include the testing code). Here you can find all the training details in our implementation.

training data sample method :

cars-196 : random sample 10 classes (each with 10 images) as a mini-batch

cub-bird : random sample 10 classes (each with 10 images) as a mini-batch

stanford-online-products : random sample 2 big classes, then sample 10 classes in each big class. (each class only have small number of images)

deep-fashion : randome sample 2 big classes, then sample 10 classes in each big class.

we will release the sample method code as soon as possile.