Im doing a object detection job, and the dataset contains many ducks
the dataset has 500 samples and I used yolov5 and retinaNet to train the model.
Now the weird thing happens, when using yolov5, my 1-shot and 5-shot mAP are like 25-30%, 10-shot is around 65%, all-dataset is around 95%.
But when using RetinaNet, my 1-shot, 5-shot, 10-shot are all aroung 80%, and the all-dataset is aroung 95%, which is pretty weird for me since I think 1-shot and 5-shot should not get such a high mAP.
I know that not showing the whole code won’t give much information about my problem, I’m just curious if anyone has similar problems? And what would probably go wrong?