The study addresses the challenge of image quality assessment in ultra-widefield fundus imaging (UWF) for diabetic retinopathy (DR) diagnosis. With the rise of UWF, which offers a broader field of view and higher resolution than traditional methods, the need for accurate image quality assessment is paramount. The research focuses on the limitations of UWF image quality assessment, which is still in its infancy compared to color fundus photography (CFP). The study utilizes a dataset of 434 UWF images, categorized as “gradable” and “ungradable” quality, to develop and validate a deep learning model. The model incorporates data augmentation and pseudo-labeling to improve generalization and robustness. By integrating various lightweight models, the study achieves a significant boost in performance, reducing inference time while securing a first-place ranking in a competition. This research underscores the importance of UWF in DR diagnosis and the necessity for robust image quality assessment to ensure accurate lesion identification and early intervention. As the top team, we have open-sourced our code on GitHub at https://github.com/yeungbo/MICCAI-UWF4DR-2024 .