I have an object detection model with 83 classes at 80 images for each class for train and at 15 images for each class in test. the model I\'m using is ssd_efficientdet_d2_768x7