Question about the training dataset
rover5056 opened this issue · 1 comments
Hi
I am a beginner at unsupervised learning and try to use swav on my own dataset. But there is a concept that makes me confused.
I was used to thinking that methods like swav do not need labeled images for the training process, it will run upon some none label images like KNN or K-MEANS cluster.
But when I run the main_swav.py file, the argument data_path needs a folder like ImageNet train. Is that mean I need to pre-label images and assign them to correspond dir to satisfy MultiCropDataset(datasets.ImageFolder). It looks like a classic supervised model training process...
Am I misunderstanding that parameter or the entire method?
Thanks for the reply and any help
Hi!
What I did is simply change the directory.
E.g. Change unlabeled -->( image1.png, image2.png, ...) to unlabeled --> temp_folder --> ( image1.png, image2.png, ...)