with Generative AI ✨, Preference Optimization ✨, Agentic Systems ✨, and World Models ✨
LoViF returns for its second edition at ECCV 2026. This workshop aims to bring together researchers and practitioners working at the intersection of low-level vision, generative models, reward learning, preference optimization, and agentic systems.
LoViF focuses on new frontiers in low-level vision driven by recent advances in generative AI, preference-aware optimization, and intelligent interactive systems. The workshop highlights both foundational research and practical systems for restoration, enhancement, compression, assessment, representation, and multimodal understanding.
We invite submissions on topics bridging low-level vision with AGI-oriented trends. The core scope of the second edition remains largely consistent with the first edition, while the exact submission format and paper page limits will follow the final ECCV 2026 workshop policy.
Low-level vision such as restoration, enhancement, compression, and perception with diffusion models, visual autoregressive modeling, GANs, and other generative priors; low-level vision for AI-generated contents including images, videos, and related media.
Explainable quality assessment and reward modeling for visual generation or processing; preference optimization methods such as DPO, GRPO, and related paradigms for perceptual alignment and multi-objective optimization.
Low-level agentic systems with LLMs or LVMs, interactive perception pipelines, human-in-the-loop optimization, reasoning-enabled restoration and enhancement, and universal framework design for adaptive visual processing.
Latent compression, token compression, quantization, sparse attention, caching, acceleration, and visual tokenizers for low-level processing, representation learning, and efficient generation or restoration systems.
Data coding, streaming, and visual processing for machine intelligence, including large visual models, autonomous driving, robotics, and UAV-related perception or downstream task adaptation.
3D/4D compression, reconstruction, generation, editing, rendering, simulation, and evaluation for world models, including Gaussian splatting, NeRF, point clouds, multimodal benchmarks, and physically grounded assessment.
Instruction tuning, prompting, zero-shot and few-shot in-context learning, multimodal reasoning, and MLLM-assisted compression, restoration, enhancement, or quality assessment.
Privacy, robustness, trustworthiness, safety for downstream tasks, and broader impacts of generative and agentic low-level vision in real-world domains such as healthcare and scientific imaging.
LoViF @ ECCV 2026 will present awards to recognize outstanding contributions from both regular workshop papers and challenge-related papers.
Presented to an outstanding regular paper accepted by LoViF @ ECCV 2026.
Presented to an outstanding challenge paper associated with LoViF @ ECCV 2026 competitions.
TBD.
The workshop dates are fixed under ECCV 2026, and the submission system is now live on OpenReview. The key dates below are based on the current workshop submission page.
ECCV 2026 workshops and tutorials.
The OpenReview submission portal is now open.
Current deadline shown on the OpenReview page.
These dates can be added once finalized.
The LoViF @ ECCV 2026 submission system is now open on OpenReview. Authors can submit their papers through the official workshop portal. The current submission deadline shown on the portal is July 10, 2026, while additional details such as acceptance notification and camera-ready deadlines may be updated later.
Submitted papers should follow the ECCV requirements. Please refer to the official OpenReview page for the latest submission instructions, formatting requirements, and review updates.
University of Science and Technology of China
Tencent
The University of Texas at Austin
Nanyang Technological University
NVIDIA
Snap Research
Kuaishou
The University of Texas at Austin
National University of Singapore
University of Science and Technology of China
University of Science and Technology of China
National University of Singapore
University of Trento
Stony Brook University
York University
York University
LoViF 2026 hosts a series of challenges and competitions to promote practical benchmarks, robust algorithms, and real-world low-level vision systems.
Our goal is to introduce a new and applicable benchmark for the development of real-world all-in-one image restoration algorithms.
This challenge introduces a new real-world benchmark for unified removal of raindrops and reflections, encouraging methods that restore clean, visually plausible backgrounds.
Our goal is to introduce a new benchmark for ultra-low bitrate image compression under extreme bitrate constraints.
Our goal is to introduce a new benchmark for the development of day and night raindrop removal for dual-focused images.
Additional competitions and challenge tracks related to low-level vision, generative restoration, enhancement, quality assessment, and real-world visual processing will be announced soon.
LoViF @ ECCV 2026 will be held in conjunction with ECCV 2026 in Malmö, Sweden.
Malmö, Sweden
ECCV 2026 runs from September 8 to 13, 2026, with workshops scheduled on September 8–9.
TBD