Hello, how to finetune own datasets

[wudizuixiaosa]

What should I do if I want to fine tune the current pre training model to my own dataset instead of Imagenet's Val dataset? Can you answer it? Thank you very much

[TeleeMa]

@wudizuixiaosa You just need to write your own Dataset Class in datasets.py. Also, don't forget change the data_path and nb_classes in args of main_finetune.py.

[wudizuixiaosa]

Thank you for your reply. I know that we can change the data_path and class in finetune..py. I've also seen dataset.py project, but I see that the settings in it are based on Imagenet. As I am a beginner, my foundation is not good, and I am very interested in your project. So can you give me an example? Let's assume that there are two kinds of dogs and cats in my dataset. How can I define them? Mainly, I don't see relevant parameter settings in the project.

[wudizuixiaosa]

By the way, which format of dataset should i choose,coco,voc or Unlabeled pictures in folders of corresponding categories？

[gaopengpjlab]

Please follow the dataset format of Imagenet. Basically, you can need to build two folders named dog and cat then put images into the corresponding folder.

[wudizuixiaosa]

Oh, thank you for replying so quickly，i am coming to try it now

[gaopengpjlab]

train
----dog
----cat
val
----dog
----cat

The folder structure is similar to the above illustration.