This repository contains the experimental implementation of Supervised Semantics-preserving Deep Hashing model
using Chainer framework. This model can generate the
semantics-preserving binary code from raw image, and it can be trained
as simple classification task.
NOTE: This is not the official implementation.
Requirements
Python
Chainer
Scipy
Jupyter (For Demo)
Matplotlib (For Demo)
Training
At first, please download the pre-trained model parameter of AlexNet. Download script is provided.
$ bash scripts/download_alexnet.sh
Then, convert the caffemodel file to npz format to save the initialization time of training script.
$ python scripts/convert_caffemodel_to_npz.py
Now, let's start training of SSDH model. Length of output binary code can be specified by --units or -u option.