Qualitative comparison of detection results on six representative positive samples from the DR10 test set. The leftmost column shows the ground-truth images with lens-system bounding boxes overlaid in green. From left to right, the remaining columns display the predictions from GL-D-FINE, D-FINE, YOLOX-LS, and Deformable-DETR. In each prediction panel, the predicted bounding box is shown in red alongside its confidence score, while the green ground-truth box is retained as a reference.