ImageFolder dataset builder for HF Hub dataset

I’ve been using load_dataset("imagefolder", data_dir="...") to load a local dataset of images and assign a series of labels from an included metadata.csv file for multi-label image classification.

However, I uploaded that dataset to the Hub and now load it in using load_dataset("hf_name",data_dir="...") but since I have to specify the Hub directory name instead of “imagefolder” for the path parameter, the labels in metadata.csv aren’t being assigned anymore.

Is there a way to fix this?


If loading from the hub, you don’t need to specify data_dir anymore.

Ah thanks for the suggestion. Still having trouble with labels being assigned using the metadata.csv though.

This is explained in detail here: Create an image dataset.

The issue was that I had an extra image in the Hub directory that wasn’t included in the metadata.csv file. Removed the image and labels are being assigned now!

This topic was automatically closed 12 hours after the last reply. New replies are no longer allowed.