LAMP-HQ: A Large-Scale Multi-pose High-Quality Database and Benchmark for NIR-VIS Face Recognition

Near-infrared-visible (NIR-VIS) heterogeneous face recognition matches NIR to corresponding VIS images. However, due the sensing gap, images often lose some identity information so that NIR-VIS issue is more difficult than conventional recognition. Recently, has attracted considerable attention in computer vision community because of its convenience and adaptability practical applications. Various deep learning-based methods have been proposed substantially increased performance, but lack training samples leads difficulty model process. In this paper, we propose a new $$\mathbf{L} {} \mathbf{a} $$ rge-Scale $$\mathbf{M} ulti- $$\mathbf{P} ose $$\mathbf{H} igh- $$\mathbf{Q} uality database ‘ $$\mathbf{LAMP}-HQ ’ containing 56,788 16,828 573 subjects with large diversities pose, illumination, attribute, scene accessory. We furnish benchmark along protocol for via generation on LAMP-HQ, including Pixel2-Pixel, CycleGAN, ADFL, PCFH, PACH. Furthermore, novel exemplar-based variational spectral network produce high-fidelity from data. A conditional module introduced reduce domain gap between data then improve performance various databases LAMP-HQ.