Advances in Image Quality Assessment and Generation

The field of image quality assessment and generation is moving towards more perceptual and efficient methods. Recent developments have focused on creating lightweight and effective metrics for image quality assessment, such as those using pseudo-MOS supervision and spatial masking. These metrics have been shown to outperform existing methods and can be used for real-time applications. Additionally, there is a growing interest in variable-scaled image generation, with methods being proposed to effectively utilize information from different aspects of diffusion models. The use of attention mechanisms and adaptive information aggregation has also been explored to improve the quality of generated images. Noteworthy papers include: MILO, which presents a lightweight perceptual quality metric for image and latent-space optimization, and InfoScale, which proposes an information-centric framework for variable-scaled image generation.

Sources

MILO: A Lightweight Perceptual Quality Metric for Image and Latent-Space Optimization

InfoScale: Unleashing Training-free Variable-scaled Image Generation via Effective Utilization of Information

VQualA 2025 Challenge on Image Super-Resolution Generated Content Quality Assessment: Methods and Results

Perception-oriented Bidirectional Attention Network for Image Super-resolution Quality Assessment

Built with on top of