I am trying to train deep learning model through jupyter notebook. I have run the road cracks detection tutorial and it works fine but when I try to train on any other dataset it gives too much validation loss and sometimes it doesn't find correct learning rate. Kindly guide what I might be doing wrong or what parameters should be set like chip_size etc before starting training ssd.
Increase epochs sizes and training set to reduce the validation loss. Please do it and let me know.
Thanks for your response. I have tried on 35 epochs recently and on 20 epochs in the past. The issue is that the validation loss is low at the beginning but it increases drastically for some epochs. Sometimes it even goes to 50000+ for some epochs. I have attached a screenshot for you. Number of training samples is 4000.
Try to increase few more training samples and epochs size increases to 70 and reduce the patch size 25%.