Unsupervised-Image-Classification
Unsupervised-Image-Classification copied to clipboard
Custom Dataset
Hi! Thanks a lot for your great work!
Can you, please, share how I can train on my custom dataset?
Want to ask the same thing, thanks!
@Cwei61 Use the same data format as datasets.ImageFolder :
|-Training Dataset ($DIR in main.sh)
|-Class#1
|-ImageFile#1 (suffix is defined as $SUFFIX in main.sh)
|-ImageFile#2
...
|-ImageFile#m
|-Class#2
...
|-Class#n
Sub-folder name is treated as ground-truth label, which is used to evaluate NMI metric with ground-truth.
If you do not have ground-truth label and just want to apply image clustering w/o performance evaluation, you can simply set only one sub-folder and put your image data into the sub-folder.