[Python] Why my training mAP in YOLOv7 is different with mAP from test.py using validation...

Im using YOLOv7 with 120 traning dataset, 60 validation dataset, dan 20 test dataset. When training, the output always print out precision, recall, mAP etc per epoch. And there is summary in the end. When im training, i got mAP 0.62. But when i tested it using test.py --task val with my validation dataset, it printed out mAP 0.87. I mean how could it be so different? Isn't it true that during training, performance testing per epoch is also done using the validation dataset? FYI im using the same confidence threshold and iou threshold when training and testing.

This is my training output

And this is when i tested it with validation dataset Testing Using Validation Dataset

Continue reading...