An Image-Based Deep Learning Approach with Improved DETR for Power Line Insulator Defect Detection

Figure 9

The training and testing loss and error of training on CPLID are plotted. The axis represents the number of training epochs and the axis denotes the value of loss or class error. Lines colored orange are outputs of the model trained from scratch, and blue lines belong to the fine-tuned model after pre-training. (a) The total loss. (b) The class error indicating the extent of misclassification. (c) The loss for class error, bounding boxes, and GIoU. The overall loss is composed of these three losses. Note that the loss curves are only for illustrating purpose, and the actual precision curves are presented in Figure 11.

(a)

(b)

(c)