Radu Timofte

Achieving state-of-the-art performance in lightweight semantic segmentation, SiConMo reveals that simplicity in design can outperform complex architectures.

Achieving a 17.9% reduction in BD-rate over VVC demonstrates that multi-scale latent representations can significantly enhance image compression efficiency.

Achieving a remarkable -16.85% BD-Rate improvement over VVC, this MoE-based approach revolutionizes learned image compression efficiency.

Ten teams achieved groundbreaking advancements in low-light image enhancement, with improvements that redefine the state-of-the-art in burst photography.

A new benchmark reveals that existing defocus deblurring methods struggle with generalization across datasets, highlighting a critical gap in current evaluation practices.

Curriculum fine-tuning can significantly enhance the success rate of LLMs in neural architecture synthesis, but distinct failure modes require different repair strategies.

Architecture-specific learning rate schedulers can boost model accuracy by over 6% compared to basic decay strategies, revealing a critical factor in neural network training success.

Source-guided neural network modifications can boost model accuracy by over 56% compared to traditional methods, demonstrating the power of LLM adaptation in model improvement.

LLMs can now generate neural architectures with 75% less code and higher accuracy by learning to write code "diffs" instead of building from scratch.

A unified benchmark reveals the trade-offs between pixel-wise accuracy and perceptual realism in state-of-the-art image super-resolution techniques.

Over 20 teams vied to decode human attention in video, revealing new insights into saliency prediction techniques.

Current image quality metrics struggle to articulate *why* one high-quality image is better than another, but this challenge shows MLLMs are closing the gap by providing expert-level explanations.

Even state-of-the-art AI-generated image detectors struggle when images are cropped, resized, or compressed, revealing a critical gap in real-world robustness.

Current image restoration models still fail to strike the right balance between noise reduction, detail fidelity, and accurate color in real-world, low-light portrait scenarios, highlighting a critical gap this challenge aims to close.

Bitstream-corrupted video restoration remains a significant challenge, even with recent advances, as revealed by the NTIRE 2026 challenge results.

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.

Forget cloud TPUs: this NAS method coaxes surprisingly good CNN architectures out of commodity GPUs using LLMs and a clever feedback loop.

Achieve state-of-the-art medical image fusion and super-resolution by jointly processing tri-modal inputs with a wavelet-guided diffusion model that explicitly handles frequency imbalances.

Achieve nearly 50% improvement in plant age and leaf count prediction by fusing CLIP embeddings with multi-view imagery, even when views are missing or unordered.