Reproduce “RT-X Net: RGB-Thermal cross attention network for Low-Light Image Enhancement” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Raman Jha, Adithya Lenka, Mani Ramanagopal, Aswin Sankaranarayanan, Kaushik Mitra
RT-X Net, a transformer-based network, integrates RGB and thermal imagery via a cross-attention mechanism to enhance low-light images. The method achieves improved image quality, sharper details, and better exposure, yielding a PSNR of 27.75 dB and SSIM of 0.85 on the LLVIP dataset, while also demonstrating robust generalization on a new real-world V-TIEE dataset.
Key insights
- Thermal images provide complementary structural and textural details, robust to low-light conditions, which can significantly improve visible light image enhancement.
- A specially designed cross-attention mechanism for RGB-thermal fusion, incorporating illumination guidance, is more effective than simpler integration strategies like channel concatenation.
- Integrating illumination features into a self-attention mechanism, inspired by Retinex decomposition, helps the network adapt to varying lighting and emphasize critical regions within feature maps.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.