How should I train my own model?

[JoeyQWu]

hello, I have a question about training a new model, and I want to know how I would train my own model, I always get the error , and I have no idea why. I would appreciate if you can tell me how to do this, thank you @elleryqueenhomels

[elleryqueenhomels]

Hi! I am not sure what error did you come across. It would be a great help if you could post the error log. Here, I guess your error comes from the "bad" images in the WikiArt. I suppose you use the script which is in the tool directory to clean up those bad images first before you train your own model. @JoeyQWu

[JoeyQWu]

@elleryqueenhomels Hi , I have just fixed the error , thanks for your kind answer , very grateful !

[JoeyQWu]

Hi, @elleryqueenhomels I just met another mistake, I trained a model, and when I transfer the images using this model, I can't get the right images, all of them are blue images or black images, I clean up the bad images using the script "preprocess.py" and all images was cleaned out, is wikiart dataset which I downloaded not correct ?

[elleryqueenhomels]

Hi! @JoeyQWu FYI I have updated this repo recently. Doing some optimization to speed up training process and inferring process. The 'preprocess' script is also been updated. Technically, if you use this latest repo to train, you will get a working model within 4 epochs (Maybe you could add some stop-early mechanism. Ex. once the loss hits a threshold, you can save the model and halt.) To your question, could please provide more info? Have you modified the network architecture ?

[elleryqueenhomels]

@JoeyQWu BTW, I suppose you use the WikiArt train set which is about 32GB and 80k images.

[395062753]

Hi@elleryqueenhomels I encountered the same problem. after training the models, the styilized image'color was not natural, please help me,Thanks!