Enhancing Text-to-Image Generation via End-Edge Collaborative Hybrid Super-Resolution

TL;DR

The paper tackles the problem of delivering high-resolution T2I outputs on resource-constrained edge servers without incurring prohibitive latency. It introduces an end-edge collaborative framework that partitions latent representations regionally and applies a region-aware hybrid super-resolution policy, using diffusion-based SR for salient foreground patches and lightweight learning-based SR for background patches, controlled by per-task choices of SR scale $S_k$ and denoising steps $D_k$ (with initial resolution $R_k=\tilde{R}_k/S_k$). A region-aware latent partitioner aligns foreground regions with SAM saliency (IoU ≈ $0.7706$), and an optimization scheme based on simulated annealing selects configurations to maximize utility $U_k=Q_k-\lambda_k T_k$ under resource constraints. Empirical results show a 33% reduction in service latency while maintaining competitive image quality, with ablations demonstrating robustness to edge capacity and the efficacy of the 25% foreground allocation; this approach offers practical gains for QoE in edge AIGC services. The framework holds significance for enabling scalable, high-fidelity T2I on edge devices, balancing perceptual quality and latency in real-world deployments.

Abstract

Artificial Intelligence-Generated Content (AIGC) has made significant strides, with high-resolution text-to-image (T2I) generation becoming increasingly critical for improving users' Quality of Experience (QoE). Although resource-constrained edge computing adequately supports fast low-resolution T2I generations, achieving high-resolution output still faces the challenge of ensuring image fidelity at the cost of latency. To address this, we first investigate the performance of super-resolution (SR) methods for image enhancement, confirming a fundamental trade-off that lightweight learning-based SR struggles to recover fine details, while diffusion-based SR achieves higher fidelity at a substantial computational cost. Motivated by these observations, we propose an end-edge collaborative generation-enhancement framework. Upon receiving a T2I generation task, the system first generates a low-resolution image based on adaptively selected denoising steps and super-resolution scales at the edge side, which is then partitioned into patches and processed by a region-aware hybrid SR policy. This policy applies a diffusion-based SR model to foreground patches for detail recovery and a lightweight learning-based SR model to background patches for efficient upscaling, ultimately stitching the enhanced ones into the high-resolution image. Experiments show that our system reduces service latency by 33% compared with baselines while maintaining competitive image quality.

Figures (8)

• Figure 1: The impact of SR scale and denoising steps. The horizontal axes indicate generation resolution (where 1000 denotes $1000\times1000$ pixels) and the number of denoising steps, respectively.
• Figure 2: Generation delay and Clipscore for different target resolutions.
• Figure 3: Visual comparison among different super-resolution strategies.
• Figure 4: Visualization of variance-based latent partitioning, where red/green patches represent high/low variance, and white/black SAM areas represent foreground/background regions.
• Figure 5: System Overview