Autonomous systems commonly rely on RGB cameras, which are susceptible to failure in low-light and adverse conditions. Infrared (IR) imaging provides a viable alternative by capturing thermal signatures independent of visible illumination. However, its high cost and integration complexities limit widespread adoption. To address these challenges, we introduce SpectraDiff, a diffusion-based framework that synthesizes realistic IR images by fusing RGB inputs with refined semantic segmentation. Through our RGB-Seg Object-Aware (RSOA) module, SpectraDiff learns object-specific IR intensities by leveraging object-aware features. The SpectraDiff architecture, featuring a novel Spectral Attention Block, enforces self-attention among semantically similar pixels while leveraging cross-attention with the original RGB to preserve high-frequency details. Extensive evaluations on FLIR, FMB, MFNet, IDD-AW, and RANUS demonstrate SpectraDiff's superior performance over existing methods, as measured by both perceptual (FID, LPIPS, DISTS) and fidelity (SSIM, SAM) metrics.

We exploit 3DGS itself to render stereo pairs and process for more accurate depth supervision. Given a camera pose among those in the training set, we derive a corresponding right viewpoint in a fictitious stereo configuration according to an arbitrary stereo baseline. During training, for each image in the training set we render a corresponding right frame; we process the two through a stereo network to obtain depth. We train 3DGS by minimizing the difference between rendered and real images, as well as between rendered depth and the depth map obtained from stereo.