Search papers, labs, and topics across Lattice.
5
0
7
0
Time conditioning, a seemingly crucial component of diffusion models like DDIM, can be entirely bypassed without sacrificing generation quality by carefully shaping the evolution of noisy data manifolds.
Forget novel architectures: massive datasets and clever training schedules can unlock surprisingly large gains in image denoising, even with established models.
Achieve state-of-the-art super-resolution by ensembling existing models *without any training*, proving that smart combination beats further architectural complexity.
Counterintuitively, leveraging CLIP to curate external data boosts nighttime image dehazing performance more effectively than complex network architectures when training data is limited.
Reconstructing 3D scenes from images obscured by smoke and extreme darkness is now significantly more achievable, thanks to insights gleaned from the NTIRE 2026 challenge.
