NTIRE 2018 Spectral Reconstruction Challenge

As part of the “New Trends in Image Restoration and Enhancement Workshop” (NTIRE 2018) we are presenting a spectral reconstruction challenge. In this challenge, participants are invited to attempt the recovery of 31 channel hyperspectral information from RGB images.

Challenge details and materials can be found here, as well as at the NTIRE 2018 website.

 

More information and code examples for the BGU reconstruction method can be found on the hyperspectral project page and the hyperspectral database page.

Challenge Tracks

Track 1: “Clean”
Recovering hyperspectral data from uncompressed 8-bit RGB images created by applying a know response function to ground truth hyperspectral information.


Track 2: “Real World”
Recovering hyperspectral data from jpg-compressed 8-bit RGB images created by applying an unknown response function to ground truth hyperspectral information.

Downloads

The BGU Natural Hyperspectral Image database has been repackaged here for your convenience.

The full training image archive (TRAIN1+TRAIN2) contains 256 hyperspectral images with 31 spectral channels from 400nm to 700nm at 10nm increments stored in .mat format.

The “NTIRE Additions” data set contains 53 images taken in close proximity (both spatially and temporally) to the NTIRE challenge validation and test image sets.

Leaderboard

Last Update: 10/01/2018

"Clean" Track

RRMSE
(Competition Metric)

MSE
(Reference Metric)

BGU SHReD (Baseline)

PCA

see the competitions codebase page for the latest results.

"Real World" Track

RRMSE
(Competition Metric)

MSE
(Reference Metric)

BGU SHReD (Baseline)

PCA

see the competitions codebase page for the latest results.