Evaluating the Impact of Online Database Integration on the Performance of a Sun Glare Detection Model: A Case Study from a Chinese High School Gym

The Rob flow sun glare dataset is integrated with the study to train the model to achieve the best performance for object detection on sun glares in a Chinese school gym. The results demonstrate that models trained with “Robflow” data have the worst performance, and models trained by combining the local and “Robflow” data perform slightly worse than those trained with local data. However, the models trained with combined data require fewer epochs to reach high performance and are more stable than those trained with local data.