RTMOSeg: Robust and Accurate Object Segmentation for Railway Tracks

In the same category, the RTMOSeg model achieves the highest test accuracy. These results demonstrate that the proposed RTMOSeg model outperforms other models in terms of segmentation effectiveness. Additionally, for different types, the test accuracy of fasteners and sleepers in other models is significantly lower compared to rails, whereas the test accuracy of RTMOSeg is only slightly lower. The experimental findings indicate that RTMOSeg is effective in segmenting both simple and complex targets. Table 3 presents the test results of each model trained on dataset 2. It is evident that RTMOSeg continues to exhibit the highest test accuracy within the same category. In multiple comparison methods, the test accuracy of fasteners and sleepers in multiple models is much lower than that in dataset 1, and the test accuracy of sleepers is lower than that of fasteners. This discrepancy arises from the increased complexity in the shape and position of fasteners and sleepers in images that contain multiple rails (i.e. images with turnouts). Compared to other models, the RTMOSeg proposed in this paper achieves the highest test accuracy, with minimal difference in test accuracy between dataset #1 and dataset #2, further confirming the strong robustness of RTMOSeg.