The challenge of non-line-of-sight (NLOS) imaging lies in the multiple reflections of light paths, causing a significant drop in signal-to-noise ratio. Visible light is easily affected by illumination conditions, making passive NLOS reconstruction algorithms based on visible light difficult to achieve clear results. However, long-wave infrared (LWIR) light provides stronger specular reflections compared to visible light, improving the signal-to-noise ratio when imaging obscured targets. While LWIR can enhance the quality of NLOS reconstructions, it typically lacks the chromatic details present in visible light. In this study, we make an unprecedented attempt to combine LWIR for high-quality reconstruction with methods to preserve color information, which is crucial for passive NLOS imaging. We introduce NLOS-I2V, an innovative end-to-end training framework. NLOS-I2V reconstructs two-dimensional images of thermal radiation captured on a relay wall and converts these blurred infrared domain images into the clear visible light domain using a generative adversarial network (GAN). This method allows for the synthesis of high-quality, long-range reconstructions while preserving color information. Extensive experiments on a custom-built LWIR NLOS dataset demonstrate exemplary performance in both quantitative metrics and subjective visual representation. The code is available at https://github.com/codeMakerZWH/NLOS-I2V .