Computer Science · Computer Vision and Pattern Recognition
GAN based Visible to Thermal Image Translation
Muhammad Abdullah, AhHyun Lee, Dongil Han
See on the mapAbstract
The increasing demands of deep learning based solutions for security and surveillance applications has opened various new challenges in computer vision. Multimodal face recognition is one of the important tasks where face recognition may be required to be performed on infrared or thermal camera images in locations with low visibility and challenging lighting conditions. However, the data available for such modalities is not enough as compared to data available for visible light images. This paper presents a generative adversarial network (GAN) based solution to solve the problem of data scarcity. The closed loop GAN system CycleGAN is used to generate novel thermal images from visible light images. The results are very impressive and close to reality.