Start training from scratch on a new data set, without calling pre-training weights.

[lwzhaojun]

Hello, thank you very much for your code. I want to start from scratch on the new data set without calling the pre-training weights. May I ask what operation is required? Is it ok to modify the data set address directly in examples/imagenet/main.py?

[AndrejHafner]

I'm also interested in fine tuning this network. I ran the main.py in examples/imagenet on my own dataset, and it starts training, but loss is not decreasing.