DualResidualAttention

Dual Residual Channel Attention Net for Spectral Reconstruction

This model has been created in the NTIRE 2022 Challenge on Spectral Reconstruction from RGB competition. We secured the 10th position in the competition. Our goal is to create a lighweight spectral reconstuction model. It can be used in mobile devices and normal computers. if we are able to get the high quality spectral images from RGB , we can find quality of the food prodcuts from mobile phone or normal laptop.

Proposed network architecture diagram

Dual residual block diagram

Environment

1. Python 3.6.4
2. Anaconda 4.9.2
3. Ubuntu 16.04 or Windows10

How to setup the environment

Step 1

Unzip the downloaded folder

Step 2

Open the powershell or terminal

Step 3

$cd yourpathtoLightWeightModel

$pwd
> ~/DualResidualAttention

$pip install --upgrade -r requirements.txt

How to test the model on your own imgaes

$python test_v2.py --testImagePath=yourpathtoimages

test results

Data size	Data	MRAE	RMSE
100	Training Data	0.73984	0.07509
50	Validation Data	0.7795	0.1016

Reference

1. "Coordinate 2D Convolution layer"
2. "LightWeightModel"