Gaussian Mixture Model Embedding

This is the Matlab implementations of our methods Gaussian Mixture Model Embedding (Gemb) accepted in ICIP17 [pdf].

The source code includes library for other hashing methods including:

• Binary Autoencoder (ba)
• Iterative Quantization (itq)
• Spectral Hashing (sh)
• Spherical Hashing (sph)
• K-meanHashing (kmh).

These libraries are slightly modified but still keeping the algorithm and recommended parameters from original papers.

BibTex

@INPROCEEDINGS{Gemb,
 author    = {Tuan Hoang and Do, Thanh-Toan and Dang-Khoa Le Tan and Cheung, Ngai-Man},
 title     = {ENHANCING FEATURE DISCRIMINATION FOR UNSUPERVISED HASHING},
 bookTitle = {ICIP},
 year      = {2017},
 month     = {Sep},
}

DATASET

The folder './dataset' contains the mat files used for this demo code. If the './dataset' folder is empty, please download the dataset here and place them in this folder.

• dataset_featureType_train.mat: The files are used for trainning and database.
  • train_features (# samples x feature_dim): Extracted features of training images
  • train_labels (# samples x 1): Semantic labels of training images
• dataset_featureType_test.mat: The files are used for testing.
  • test_features (# samples x feature_dim): Extracted features of testing images
  • test_labels (# samples x 1): Semantic labels of testing images

Where:

• dataset: cifar10/mnist/labelme
• featureType: gist/vggfc7 (for mnist dataset, only gist feature is available)

USAGE

Modify the parameters which are defined and explained in 'main_demo.m' properly then run.