DRSH
Deep Recurrent Scaling Hashing
Code for paper Supervised Hashing with Recurrent Scaling. Apperas in 3rd Asia Pacific Web and Web-Age Information Management Joint Conference on Web and Big Data (APWeb-WAIM), 2019.
Requirements
- theano
- keras
- python 2.7
Run the codes
To run the training/test codes you should download the raw images of NUS-WIDE dataset here first, and put the images in one folder. Then using the trained VGG-16 caffemodel to extract features. The attributes are extracted through the finetuned caffemodel training on attributes.
The finetuned caffemodel for extracting attributes can be downloaded here. The file that contains labels of attributes of COCO dataset for extracting NUS-WIDE images' attributes can be downloaded here. The classified h5 file for finding positive and negative triplets can be downloaded here. You also should download the 300-dimension GloVe vectors.
After getting the files mentioned above, run the NUSex_att.py, e_x.py and transfer.py to get NUSGlove.h5. Run the NUSex_fea.py to get NUSF.h5.
Models
The models for generating hash codes in different code lengths can be downloaded here
Reference
If you find this codebase useful in your research, please consider citing the following paper:
@InProceedings{Fu2019DRSH,
author = {Fu, Xiyao and Bin, Yi and Wang, Zheng and Wei, Qin and Chen, Si},
title = {Supervised Hashing with Recurrent Scaling},
booktitle = {The Asia Pacific Web (APWeb) and Web-Age Information Management (WAIM) Joint Conference on Web and Big Data(APWeb-WAIM)},
month = {August},
year = {2019}
}