Problem
Research questions and friction points this paper is trying to address.
Large Image Super Resolution
Dynamic Network
Patch Difficulty Assessment
Quality-Complexity Trade-off
ENAF: A Multi-Exit Network with an Adaptive Patch Fusion for Large Image Super Resolution
🤖 AI Summary
This study addresses the weak correlation between handcrafted evaluation metrics and computational complexity in large-scale image super-resolution by proposing ENAF, a dynamic network architecture. The method introduces a lightweight PSNR estimation network to replace manual scoring, accurately modeling the mapping between texture features and computational costs. By integrating multiple early-exit mechanisms with adaptive patch fusion, ENAF enables dynamic allocation of computational resources. Experimental results demonstrate that ENAF significantly optimizes the quality-complexity trade-off across various settings, effectively mitigating efficiency bottlenecks in high-resolution image super-resolution. Consequently, this work establishes a novel paradigm for adaptive inference, offering a robust solution for balancing reconstruction fidelity with computational demands in resource-constrained scenarios.
📝 Abstract
To accelerate single image super-resolution (SISR) networks on large images (2K-8K), many recent approaches decompose an image into small patches and dynamically determine an execution path according to its difficulty (referred to as a dynamic network). To quantify the hardness of a patch, they mainly rely on a handcrafted assessment score, e.g., edge, which weakly associates a patch's texture with the computational complexity of a SISR model. To address the problem, we introduce ENAF - a dynamic network for SISR with an adaptive patch fusion. Built on top of a backbone, ENAF incorporates multiple early exits (EEs) to tackle the over-parameterized SISR model. More importantly, ENAF plugs a tiny network that estimates PSNR to associate data texture with a computation cost at an EE. Based on the scores, ENAF effectively assigns image patches to an exit, enhancing the quality-complexity trade-off. Extensive experiments on common datasets with popular SISR backbones demonstrate the effectiveness of ENAF in various settings. The source code is provided in https://github.com/nmduonggg/ENAF
Innovation
Methods, ideas, or system contributions that make the work stand out.
Adaptive Patch Fusion
Multi-Exit Network
Dynamic SISR
PSNR Estimation
Early Exit
🔎 Similar Papers
No similar papers found.
