Tuesday 11 March 2025
The quest for accurate image quality assessment has been a long-standing challenge in the field of computer vision. Researchers have been working tirelessly to develop models that can accurately predict the quality of images, taking into account various factors such as distortion, noise, and compression. In recent years, large language models (LLMs) have shown promising results in this regard.
The latest development comes from a team of researchers who have proposed a novel approach to image quality assessment using LLMs. Their method, dubbed DeQA-Score, leverages the power of LLMs to predict not only the overall quality score but also the distribution of scores for each image. This is achieved by discretizing the score distribution into soft labels, which allows for more nuanced and accurate assessments.
The team evaluated their approach on multiple benchmarks, including the widely used KonIQ, SPAQ, KADID, and LIVE-Wild datasets. The results were impressive, with DeQA-Score outperforming state-of-the-art methods in terms of accuracy and robustness. Moreover, the model was able to predict the score distribution closely aligned with human annotations.
The researchers also explored the potential benefits of co-training LLMs for image quality assessment with language-based quality description tasks. While this approach did not yield significant improvements in score regression performance, it did demonstrate that high-level quality description datasets can enhance low-level perception tasks.
One limitation of DeQA-Score is its tendency to introduce errors when dealing with images having extremely small variance. However, the researchers acknowledged that such cases are rare and the overall impact on accuracy is relatively minor.
The implications of this research are far-reaching. With the ability to accurately assess image quality, applications can be developed that automatically adjust compression rates, noise reduction algorithms, or even optimize image transmission protocols. Additionally, DeQA-Score has the potential to improve the performance of various computer vision tasks, such as object detection and segmentation.
The authors’ approach is a significant step forward in the quest for accurate image quality assessment. By leveraging the power of LLMs and discretizing score distributions into soft labels, they have developed a model that can accurately predict not only overall quality scores but also nuanced aspects of image quality. As the field continues to evolve, it will be exciting to see how DeQA-Score is applied in various applications and how future research builds upon this innovative approach.
Cite this article: “Accurate Image Quality Assessment with Large Language Models”, The Science Archive, 2025.
Computer Vision, Image Quality Assessment, Large Language Models, Deqa-Score, Accuracy, Robustness, Human Annotations, Score Regression, Compression Rates, Object Detection
