PDF

Bibtex

@Article{s17040873,
AUTHOR = {Aguilera, Cristhian A. and Sappa, Angel D. and Aguilera, Cristhian and Toledo, Ricardo},
TITLE = {Cross-Spectral Local Descriptors via Quadruplet Network},
JOURNAL = {Sensors},
VOLUME = {17},
YEAR = {2017},
NUMBER = {4},
ARTICLE NUMBER = {873},
URL = {http://www.mdpi.com/1424-8220/17/4/873},
ISSN = {1424-8220},
DOI = {10.3390/s17040873}
}

First install the torch framework and cudnn

1. Install torch
2. Cudnn torch

Two options: Download the generated t7 dataset from

• Google Drive

Training and evaluation t7 files are different

1. Download the dataset from

• Google Drive

Evaluation code can be found in the eval folder. To eval one sequence:

1. You have to generate or download the nirscenes patch dataset
2. Install xlua

luarocks install xlua
luarocks install moses

3. Run

cd eval
th nirscenes_eval.lua -dataset_path [path] -net [trained network]

For example, to eval the field sequence using the Q-Net article trained network.

th nirscenes_eval.lua -dataset_path ../datasets/nirscenes/test -net ../trained_networks/qnet.t7

For more options, run

th nirscenes_eval -h

1. You have to download the dataset first
2. Run

cd eval
th icip2015_eval.lua -dataset_path ../datasets/icip2015/ -net [trained network] 

For example. To eval Q-Net

cd eval
th icip2015_eval.lua -dataset_path ../datasets/icip2015/ -net ../trained_networks/qnet.t7 

1. Install penlight, torchx and json

luarocks install penlight
luarocks install torchx
luarocks install json

2. Train a network

cd train
th nirscenes_quadruplets_train.lua

Run

th nirscenes_quadruplets_train.lua -h

to see the options

Note The training code is different from the one used in the article. This new version is smaller. Additionally, the dataset was generated from zero. So, small differences in FPR95 may happen.