LoViF Workshop @ CVPR 2026

Call for Papers

We invite submissions on topics bridging low-level vision with AGI trends. The submission follows the template of CVPR2026 and does not exceed 8 pages (not including references)

Generative Models for Visual Processing

Low-level vision (e.g. restoration, enhancement, compression, perception) with visual autoregressive modeling (VAR), diffusion models, and GAN...; Low-level vision for AI-generated contents (AIGC images/videos and others);...

Reward Models & Preference Optimization in Visual Processing

Explainable quality assessment and reward models for generation or processing; Preference optimization (e.g., DPO, GRPO) for visual processing and perception; ...

Agentic Systems for Interactive Low-Level Vision

Human-machine interaction theory in visual processing; Low-level agentic systems with LLMs/LVMs; Universal framework design; ...

Compact and Informative Low-level Representation

Latent compression/Cache/Acceleration/Quantization/sparse attention for representation, visual signal generation or low-level processing; Low-level token compression and visual tokenizer; ...

Low-level Vision for Machine Tasks

Data coding, streaming, and visual processing for machine and intelligent tasks, such as large foundation models (LVMs), autonomous vehicles, robotics, or UAVs...

Low-level Vision in World Models

3D/4D (gaussian splatting, Nerf, point cloud...) compression, reconstruction/generation, audio-visual benchmark, editing, gaussian splatting, world model evaluation, rendering \& simulation pipelines...

LLMs & Multi-modal LLMs for Low-level Tasks

Multi-modal low-level visual signal processing, compression and related instruction tuning, prompt techniques and zero/few-shot in-context learning...

Ethics & Social Impact for Generative Low-level Vision

Ethical and impacts of generative low-level vision, including privacy in agentic systems, safety and robustness for downstream tasks, and impacts in healthcare and biomedical imaging...

Invited Speakers

Dimitris Samaras

Stony Brook University

Yiyi Liao

ZJU

Kangfu Mei

Google DeepMind

Hunyuan

Tencent

Michael S. Brown

York Univ.

To Be Determined

We have invited some experts in generative models and low-level vision from school and industry.
Please check back soon for updates.

LoViF 2026 Challenges

Participate in our challenge tracks.

CHALLENGE START Feb 6, 2026

01

The First Challenge on Real-World All-in-One Image Restoration

Establishing a new benchmark for complex real-world degradations, this challenge evaluates all-in-one restoration through both quantitative metrics and visual quality.

Organization: EntroVision Join Competition

02

The Challenge on Efficient VLM for Multimodal Creative Quality Scoring

Tasked with predicting product ratings from images, videos, and text, this challenge focuses on developing lightweight, high-performance VLMs for real-world creative quality assessment.

Organization: Snap Inc. & NTU & SYSU Join Competition

03

The Challenge on Weather Removal in Videos

This challenge focuses on video restoration under complex weather conditions, emphasizing temporal consistency and robustness through the integration of physical priors and temporal reasoning.

Organization: Leeds & Tencent & USTC Join Competition