TOC-SR: Task-Optimal Compact diffusion for Image Super Resolution

2026-05-04 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision & Pattern Recognition · Depth: Expert, quick

Summary

TOC-SR is a novel framework designed to create efficient, one-step super-resolution models by first identifying a compact diffusion backbone. This approach begins with a sixteen-channel latent diffusion model and employs feature-wise generative distillation to construct parameter-efficient surrogate blocks. Architecture discovery is then performed using epsilon-constrained Bayesian Optimization, aiming to minimize model complexity while preserving generative fidelity. The resulting compact diffusion backbone achieves a significant 6.6x reduction in parameters and a 2.8x reduction in GMACs compared to the expanded diffusion model. This backbone is subsequently adapted for super-resolution, and the diffusion process is distilled into a single-step generator, demonstrating efficient super-resolution with strong reconstruction quality.

Key takeaway

For research scientists developing image restoration models, TOC-SR offers a pathway to significantly reduce computational costs and model size without sacrificing reconstruction quality. You should consider applying generative distillation and Bayesian Optimization for architecture discovery to optimize diffusion models for practical deployment, especially for super-resolution tasks where efficiency is critical.

Key insights

TOC-SR creates efficient one-step super-resolution models via compact diffusion backbone discovery and distillation.

Principles

• Minimize complexity while preserving fidelity.
• Distill multi-step processes into single-step generators.

Method

Construct parameter-efficient surrogate blocks via feature-wise generative distillation, then use epsilon-constrained Bayesian Optimization for architecture discovery, and finally distill the diffusion process into a single-step generator.

In practice

• Reduce model parameters by 6.6x.
• Achieve 2.8x reduction in GMACs.

Topics

• TOC-SR
• Diffusion Models
• Image Super-Resolution
• Model Compression
• Generative Distillation

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

Related on AIssential

• teasr: training-efficient any-step diffusion transformer for real-world image super-resolution — TEASR uses self-adversarial distillation within a single diffusion model for efficient, flexible Real-ISR, eliminating auxiliary teachers.
• The Fourth Challenge on Image Super-Resolution ($\times$4) at NTIRE 2026: Benchmark Results and Method Overview — The NTIRE 2026 SR challenge benchmarks $\times$4 image super-resolution across fidelity and perceptual quality.
• Image Restoration via Diffusion Models with Dynamic Resolution — Dynamic resolution DMs accelerate image restoration by operating in lower-dimensional subspaces, avoiding VAE overhead.
• A Scale-Adaptive Framework for Joint Spatiotemporal Super-Resolution with Diffusion Models — A scale-adaptive framework reuses one architecture for spatiotemporal super-resolution across diverse scaling factors.
• AsyncPatch Diffusion: spatially-flexible image generation — AsyncPatch Diffusion enables spatially flexible image generation by assigning distinct noise levels to different regions, unifying various generative tasks.
• Adapting Diffusion Language Models for Lossless Pixel-Level Image Transmission — DDM-SSCC adapts diffusion language models for lossless pixel-level image transmission, achieving superior exact recovery via novel coding and denoising strategies.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.