Ultra-widefield fundus images provide a broad view and play an important role in the integration of deep learning and healthcare. Therefore, it is important to obtain high-quality ultra-widefield fundus image data. We participated in Task 1 of the Ultra-Widefield Fundus Imaging for Diabetic Retinopathy Challenge, focusing on ultra-widefield fundus image quality assessment. The performance of the image quality assessment can be improved by tricks in the training and inference procedure, such as data augmentation, label smoothing, image resizing, and integration of deep learning models. We employ the bag of tricks to enhance the performance of ultra-widefield fundus image quality assessment. In this paper, we examine a series of such tricks and empirically assess their impact on the final model through experiments. The experiments demonstrate that by combining these improvements, significant improvements in prediction performance can be achieved. We achieve a test score of 0.9644 in the image quality assessment task of the challenge.