How can I use my own dataset to train the model?

[wushilian]

I want to use the simgan to train other dataset，so what format does the data need？

[mjdietzx]

Uses Keras image preprocessing. Look at keras documentation but this is pretty simple. Expects each dataset to be in a directory containing directories of images corresponding to label.

[dagoPA]

So in order to use a custom dataset what code file do i have to modify? Or it is just to have the right directory structure?

[mjdietzx]

Hey, you shouldn't have to modify the code. Synthesized images in one dir, refined in another. Make sure directories are named correctly. See keras image processing docs for more info.

[dagoPA]

Another question, Once the network is trained, how do i generate refined images?

[wushilian]

@dagoPA Did you solve this problem?how to get refined images?

[dagoPA]

@wushilian I haven't yet Because i think this project only takes a random small set of data to inference and instead of saving it only saves the weights and the plot

[wushilian]

@dagoPA It seems like that.

[dagoPA]

@wushilian i think in the code i stead of predict batch we need a loop that iterates predicting each image in target directory

[wushilian]

@dagoPA that's all right.And I think cycle GAN and simGAN are very similar,so now I use cycleGAN to Converte synthetic data into real data

[dagoPA]

@wushilian I saw you forked that project, im new to github but i hope i can know your email

[wushilian]

@dagoPA I guess you have saw my email,so I deleted it.