Training SSD with ResNet50 gives low accuracy

[sbastani]

You have mentioned in the readme that you obtained an accuracy of 0.756. I have done training with the same configs for about 120 epochs but gets accuracy of 0.59. I trained from scratch (no pre-trained base network). I was wondering if you used pre-trained ResNet, and how many epochs did you train. Did you also do fine tuning. It would be very good if you elaborate more on how you trained the model.