notesum.ai
Published at November 25CapHDR2IR: Caption-Driven Transfer from Visible Light to Infrared Domain
cs.CV
Released Date: November 25, 2024
Authors: Jingchao Peng1, Thomas Bashford-Rogers1, Zhuang Shao2, Haitao Zhao3, Aru Ranjan Singh1, Abhishek Goswami1, Kurt Debattista1
Aff.: 1WMG, University of Warwick; 2School of Engineering, Newcastle University; 3School of Information Science and Technology, East China University of Science and Technology

| Input | Methods | PSNR (×10) | SSIM (×0.1) | MSE (×0.1) | LPIPS (×0.1) |
|---|---|---|---|---|---|
| SDR | Histogram Matching (Castleman 1996; Gonzalez and Woods 2008) | 0.973 | 2.073 | 6.221 | 4.851 |
| Color Transfer (Reinhard et al. 2001) | 1.134 | 4.150 | 5.814 | 5.632 | |
| CycleGAN (Zhu et al. 2017) | 1.082 | 5.098 | 4.149 | 4.689 | |
| Pix2Pix (Isola et al. 2017) | 1.072 | 5.041 | 4.041 | 4.649 | |
| MUNIT (Huang et al. 2018) | 1.409 | 4.578 | 3.886 | 5.823 | |
| sRGB-TIR (Lee et al. 2023) | 1.202 | 3.691 | 7.858 | 4.624 | |
| RGB2IR (Huang, Huang, and Wu 2024) | 1.645 | 5.152 | 3.291 | 4.327 | |
| CapHDR2IR (Ours) | 1.879 | 6.163 | 2.574 | 3.141 | |
| HDR | Histogram Matching (Castleman 1996; Gonzalez and Woods 2008) | 1.023 | 2.174 | 6.138 | 4.613 |
| Color Transfer (Reinhard et al. 2001) | 1.243 | 4.147 | 5.003 | 5.419 | |
| CycleGAN (Zhu et al. 2017) | 1.057 | 5.151 | 4.094 | 4.547 | |
| Pix2Pix (Isola et al. 2017) | 1.066 | 5.138 | 4.060 | 4.730 | |
| MUNIT (Huang et al. 2018) | 1.386 | 4.889 | 3.913 | 5.914 | |
| sRGB-TIR (Lee et al. 2023) | 1.495 | 5.019 | 3.332 | 4.528 | |
| RGB2IR (Huang, Huang, and Wu 2024) | 1.750 | 5.530 | 2.756 | 4.066 | |
| CapHDR2IR (Ours) | 1.976 | 6.359 | 2.242 | 3.035 |