YOLOX model weights were saved every 10 epochs during training to enable resumption in case of interruptions. The convergence of total training loss for each model, shown in Figures 5c and 5f, indicates that YOLOX-nano and YOLOX-s reach loss values of 1 and 2.5 after 50,000 and 28,000 training iterations, respectively (Figures 5 c and f). The sudden increase in total loss observed in the YOLOX-s training process at step 15,000 (Figure 5f) resulted from resuming training with saved weights after an interruption.