This code is a re-implementation of the following paper: Mengran Gou, Fei Xiong, Octavia Camps, Mario Sznaier. "MoNet: Moment embedding network", CVPR 2018

The code is tested on Ubuntu 14.04 using a single NVIDIA Titan X GPU, 64G RAM and MATLAB R2016b.

Dependency

• VLFEAT
• MatConvNet
• VGG-VG16 model

How to run

• Download the pretrained VGG model and put it under ./data/models
• Download one dataset from the following list and put it under ./data
• Modify the paths in './base/setup.m'
• Run "script_run_exp.m". The default setting will train/test MoNet on CUB

Supported datasets

• CUB-200-2011
• FGVC aircraft
• Stanford cars

Known issues

• If you kill the code during the pre-trained feature extraction stage, you have to delet ./data/%NETWORK%/%DATASET%/nonftGlbGau and then re-start

Acknowledgement

This code is modified from the G2DeNet package and based on BCNN code. Please refer to the originial websites for more details

• G2DeNet
• BCNN