EvTexture++: Event-Driven Texture Enhancement for Video Super-Resolution

2026-06-12 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Emerging Technologies & Innovation · Depth: Expert, extended

Summary

EvTexture++ is an event-driven framework designed for texture enhancement in video super-resolution (VSR), utilizing event camera signals for high-frequency spatiotemporal details. It introduces a dedicated texture enhancement branch with an Iterative Texture Enhancement (ITE) module that progressively refines textures using high-temporal-resolution events. Additionally, a Temporal Texture Alignment (TTA) module enhances inter-frame consistency by integrating event-guided texture-aware flow with RGB-based flow, mitigating texture flickering. EvTexture++ operates as a flexible plug-and-play module, seamlessly integrating into existing VSR models like IART, yielding significant performance improvements, such as gains of up to 1.55 dB in PSNR on the texture-rich Vid4 dataset. The framework achieves leading performance with 10.15 million parameters and 808.6 GFLOPs per frame, demonstrating a favorable trade-off between efficiency and quality. Training takes approximately 4 days for the standalone model and 6 days for plug-in variants on four NVIDIA RTX 3090 GPUs.

Key takeaway

For Machine Learning Engineers developing video super-resolution systems, you should consider integrating event-driven modules to significantly enhance texture recovery and temporal consistency. EvTexture++ demonstrates that utilizing high-frequency event signals can boost PSNR by over 1.5 dB on texture-rich datasets, even when used as a plug-in with existing backbones. This approach offers a computationally efficient way to improve visual quality, especially in scenarios with complex textures or large motions, without requiring extensive retraining of your core VSR model.

Key insights

Event signals offer high-frequency details crucial for texture restoration and temporal consistency in video super-resolution.

Principles

• Events enhance texture recovery in VSR.
• Iterative refinement improves detail extraction.
• Dual-stream flow aligns textures robustly.

Method

EvTexture++ uses an ITE module for iterative texture refinement from event voxel grids and a TTA module for event-guided, texture-aware motion alignment, fusing event and RGB cues.

In practice

• Integrate event cameras for VSR texture.
• Employ iterative refinement for detail.
• Combine event and RGB flow for alignment.

Topics

• Video Super-Resolution
• Event-Based Vision
• Texture Enhancement
• Temporal Consistency
• Deep Learning Models
• Plug-in Architectures

Code references

• DachunKai/EvTexture

Best for: Research Scientist, AI Scientist, Computer Vision Engineer, Machine Learning Engineer

Related on AIssential

• EvTexture++: Event-Driven Texture Enhancement for Video Super-Resolution — EvTexture++ enhances VSR texture and temporal consistency by leveraging event signals beyond motion refinement.
• FS-DVS: A Frequency-Selective Dynamic Visual Sensing Paradigm for Enhancing Information Completeness — FS-DVS uses a learnable pre-trigger spatial filter to mimic RGCs, enhancing event camera data completeness and noise resilience.
• HYDRA-X: Native Unified Multimodal Models with Holistic Visual Tokenizers — Unifying image and video tokenization in a single ViT with specific architectural choices significantly enhances multimodal model performance and editing consistency.
• Transformer-Based Inpainting for Real-Time 3D Streaming in Sparse Multi-Camera Setups — A transformer-based inpainting method enhances real-time 3D streaming quality in sparse multi-camera AR/VR setups.
• Beyond MMSE: Enhancing PnP Restoration with ProxiMAP — ProxiMAP enhances PnP image restoration by aligning denoiser noise schedules to prevent "cartoon-like" artifacts from inexact learned scores.
• EPRBench: A High-Quality Benchmark Dataset for Event Stream Based Visual Place Recognition — Event cameras and LLMs enhance Visual Place Recognition robustness and interpretability in challenging environments.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.