Yolov8 model training failing to match ground truth labels on objects during training

I am working on fruit detection in cucumber plant images from the greenhouse. I export my annotated dataset to colab to train on yolov8. During training after checking the first training images, i notice that ground truth labels are out of target and after training, i get a an precision of 40.8%,mAP of 17.1% and recall of 21.2%. Inference on training data gave bad results.