Due to the advantage that a thermal camera robustly works in various illumination conditions, it has become a crucial sensor in real-world applications, such as self-driving, advanced driver assistance as well as a surveillance system. However, unlike RGB images with color information, thermal images do not contain abundant information. This disadvantage makes it users difficult to recognize thermal scenes. In this paper, we aim to make pseudo-RGB that can be used at day and night by receiving thermal images as inputs and to show the necessity of pseudo-RGB research through colorization that synthesizes chromaticity of RGB in thermal images. Furthermore, we evaluate models to see whether the produced colorized image can be interpreted by perceptual models for tasks, such as pedestrian detection and depth estimation by feeding colorized images to the models that are trained with original RGB images. These experiments explicitly show the limitation and possibility of thermal colorization.