Research map / Image generation / Super-resolution
LaMAR: Laplacian Pyramid for Multimodal Adaptive Super Resolution (Student Abstract)
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes LaMAR, a lightweight (612K parameters) Laplacian-pyramid-guided model that super-resolves low-resolution thermal images using a high-resolution RGB guide, trained with MSE plus GAN loss.
Problem
Guided thermal super-resolution: enhancing low-resolution thermal images using RGB guidance, for both aligned and misaligned inputs, while avoiding the large parameter counts and slow inference of prior fusion models.
Approach
Decompose the high-resolution grayscale RGB image into a Laplacian pyramid of depth 2, replace the residual with the low-resolution thermal image, and fuse the two in a two-branch residual CNN so that no high-resolution feature maps are computed.
Key results
- On ULB17-VT, LaMAR reaches 33.635 dB PSNR and 0.9102 SSIM, above CMSR (29.928 dB, 0.882).
- On FLIR-ADAS, LaMAR reports 30.68 dB PSNR, above PAG-SR (29.56 dB), but its SSIM (0.701) is lower than the baselines (PAG-SR 0.912).
- The model has 612K trainable parameters and is reported to have faster inference than prior fusion models.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use LaMAR instead of CMSR | Guided thermal super-resolution · ULB17-VT | PSNR (dB): 29.928 -> 33.635 | Helped |
| Use LaMAR instead of CMSR | Guided thermal super-resolution · ULB17-VT | SSIM: 0.882 -> 0.9102 | Helped |
| Use LaMAR instead of PAG-SR | Guided thermal super-resolution · FLIR-ADAS | PSNR (dB): 29.56 -> 30.68 | Helped |
| Use LaMAR instead of PAG-SR | Guided thermal super-resolution · FLIR-ADAS | SSIM: 0.912 -> 0.701 | Hurt |
Surprising findings
- On FLIR-ADAS, the PSNR gain comes with a markedly lower SSIM (0.701 vs 0.912 for PAG-SR), which the authors attribute to night imagery where Laplacian pyramids lack edge guidance.
Limitations
- Weaker perceptual quality and SSIM on night-time FLIR-ADAS imagery, where Laplacian pyramids lack edge guidance.
- Results are reported only on two datasets and the abstract gives no numeric inference-time comparison.
- Loss weighting and training details are sparse (the GAN loss weight is not specified).
Experimental setup
Datasets: ULB17-VT FLIR-ADAS
Benchmarks: ULB17-VT FLIR-ADAS
Metrics: PSNR SSIM
Baselines: VTSRGAN VTSRCNN CMSR PixTransform Deep-ISTA PAG-SR
Builds on: LapSRN SRCNN SRGAN CoReFusion
Closest papers by meaning
- AdaBM: On-the-Fly Adaptive Bit Mapping for Image Super-Resolution (CVPR 2024 · similarity 0.86)
- NSARM: Next-Scale Autoregressive Modeling for Robust Real-World Image Super-Resolution (NeurIPS 2026 · similarity 0.85)
- Edge-SD-SR: Low Latency and Parameter Efficient On-device Super-Resolution with Stable Diffusion via Bidirectional Conditioning (CVPR 2025 · similarity 0.85)
- Noise-Started One-Step Image Super-Resolution via LR-Conditioned SplitMeanFlow and GAN Refinement (NeurIPS 2026 · similarity 0.85)
- Auto-Encoded Supervision for Perceptual Image Super-Resolution (CVPR 2025 · similarity 0.85)
- BUFF: Bayesian Uncertainty Guided Diffusion Probabilistic Model for Single Image Super-Resolution (AAAI 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.