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 |