fjr-nancy / joint-cluster-cnn

A tensorflow version of JULE (Joint Unsupervised Learning of Deep Representations and Image Clusters).

joint-cluster-cnn

An unofficial implementation of JULE (Joint Unsupervised Learning of Deep Representations and Image Clusters) in TensorFlow.

File Structure

Main class: joint_cluster_cnn.py

Test file: test.py

Results

Dataset	MNIST-test	USPS	COIL-20	COIL-100	UMist
JC-NL	0.849	0.883	0.904	-	0.750
JC-SF	0.861/0.876	0.875/0.858	0.967/1.000	0.978	0.840/0.880
JC-RC	0.900/0.915	0.944/0.913	0.983/1.000	0.985	0.849/0.877
OURS-NL	0.841	0.911	0.933	0.957	0.767
OURS-SF	0.859	0.901	0.956	0.964	0.945
OURS-RC	0.888	0.921	0.964	0.915	0.941

joint-cluster-cnn's Issues

Any example about the 'self-contained cluster' issue?

joint-cluster-cnn/joint_cluster_cnn.py

Line 271 in 200274b

# Assert whether there are some self-contained clusters

训练不生成权重吗

如果我用大量图片训练完了之后，要对一个图片进行分类呢

请问，这个代码运行结果与论文给出的结果一致吗？

Is this a TF implementation for JULE?

Is this a TF implementation for JULE (https://arxiv.org/abs/1604.03628)?

I'm searching for Tensorflow version for this.

Thanks.

聚类和卷积

您好！
我想问一下，这个基于无监督学习的卷积网络有相关的论文参考么，因为最近看了一篇论文实现思路与您的代码很想。