Background

About ICASSP

The IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP) is one of the world’s most renowned conferences in the field of signal processing. Organized annually by the IEEE Signal Processing Society, ICASSP brings together researchers, practitioners, and academics from around the globe to present their latest research findings and innovations.

The conference covers a broad range of topics, including audio and speech processing, image and video processing, machine learning, bioinformatics, and many other areas related to signal processing. ICASSP serves as a platform for experts to exchange ideas, foster collaboration, and contribute to the advancement of the signal processing field.

GC-5: Hyperspectral Skin Vision Challenge

As part of ICASSP 2024, the GC-5: Hyperspectral Skin Vision Challenge was introduced, focusing on the reconstruction of skin spectral reflectance in both the visible (VIS) and near-infrared (NIR) spectral ranges. This grand challenge is designed to push the boundaries of what can be achieved with hyperspectral imaging, specifically targeting applications in the cosmetic and beauty industries.

The challenge involves reconstructing hyperspectral information from RGB images captured by everyday cameras, alongside NIR data at 960nm. The goal is to make rich hyperspectral information accessible on consumer devices, paving the way for personalized beauty and skincare solutions. This competition is a significant step towards democratizing skin analysis technology, making it available directly on consumer devices like smartphones.

The GC-5 challenge is organized by leading experts in the field, including Konstantinos Plataniotis, Juwei Lu, Pai Chet Ng, and Zhixiang Chi. Their efforts aim to bring together researchers, machine learning experts, and cosmetic professionals to advance the field of beauty technology through this competition.

For more detailed information about the challenge, you can visit the official challenge website.

---

Dataset

The dataset used for the challenge is a carefully selected subset of the comprehensive data collected for the NeurIPS 2023 project. It includes hyperspectral images alongside corresponding RGB and NIR (960nm) images, specifically chosen to support the objectives of this competition.

Task Description

The main task of the ICASSP-HS2024 Grand Challenge was to reconstruct hyperspectral images from the given RGB+NIR (960nm) data. Participants were tasked with developing algorithms capable of generating high-quality hyperspectral reconstructions, which were then evaluated against a ground truth dataset.

Competition Results

Leaderboard

The metric used is the Spectral Angle Mapper (SAM), which assesses the reconstructed hyperspectral images, ensuring fidelity and accuracy, in line with the competition goals. Below is the leaderboard showcasing the top 5 teams in the ICASSP-HS2024 Grand Challenge:

Citation

If you used the dataset for your publication, kindly acknowledge and cite the following work:

@INPROCEEDINGS{10626113,
author={Ng, Pai Chet and Chi, Zhixiang and Low, Malcolm and Lu, Juwei and Plataniotis, Konstantinos N and Boulgouris, Nikolaos and Bourlai, Thirimachos and Ro, Yong Man},
booktitle={2024 IEEE International Conference on Acoustics, Speech, and Signal Processing Workshops (ICASSPW)},
title={Hyperspectral Skin Vision Challenge: Can Your Camera See Beyond Your Skin?},
year={2024},
volume={},
number={},
pages={59-60},
keywords={Reflectivity;Signal processing;Reconstruction algorithms;Propulsion;Cameras;Skin;Product development;hyperspectral skin vision;skin spectral reconstruction},
doi={10.1109/ICASSPW62465.2024.10626113}}