1. Precision和Recall
Calculation formula:
Among them, TP is the number of cases correctly identified, and FN is the number of cases with targets but identified as no targets. FP is the number of cases that have no targets but are identified as having targets.
The meaning of Precision is the ratio of the target cases correctly identified by the model to all the target cases considered by the model;
The meaning of Recall is the proportion of target cases correctly identified by the model to all correct cases.
2. mAP_0.5 and mAP_0.5:0.95
mAP refers to the average accuracy rate. In object detection, each category can draw a curve according to recall and precision, AP is the area under the curve, and mAP is the average value of multiple categories AP.
mAP_0.5 refers to the mAP when the loss function IoU is 0.5, and mAP_0.5:0.95 refers to the average value of mAP when the IoU is 0.5~0.95 (step 0.05).
The YOLO series algorithm uses the IoU loss function to measure the degree of overlap between the detection frame and the real frame. A commonly used calculation method for the loss function GIoU is:
Among them, A and B represent any two boxes, C represents the shape including AB, and IoU represents the similarity of two arbitrary frames.
3. loss
The loss function refers to the error of a single example
cls_loss represents the loss of confidence, box_loss represents the loss of the predicted frame position, and obj_loss represents the loss of the target.