Keqiang Li

GLFS achieves superior low-light image enhancement by leveraging physical priors, outperforming existing methods in both illumination correction and detail preservation.

AIGS-Net achieves superior low-light image enhancement with only 40 learnable parameters, redefining the trade-off between quality and efficiency in image processing.

Achieving high-fidelity low-light image enhancement without the common pitfalls of color distortion and structural artifacts could redefine standards in vision tasks.

Achieving state-of-the-art dehazing performance with a zero-shot approach, this method leverages 2D Gaussian Splatting to redefine how we model hazy images.

Fi-Gaussian recovers fine details in dehazed images by decoupling low and high-frequency information, outperforming traditional methods that struggle with detail preservation.

Autonomous vehicles can now better predict those rare, but critical, "edge case" maneuvers thanks to a new method that focuses on learning from the most challenging and risky scenarios.