Dataset Question

I have a collection of garment images (shirts, pants, shoes, etc.). I want to classify them in a similar manner to the dogs and cats challenge. I can put them in the directory structure such as

pants
– 1.jpg
– 2.jpg
shirt
– 3.jpg
– 4.jpg

Some items will be cross-category though. For instance, a garment might be considered a shirt and pants (not a good example, but still).

If an image item can be part of two categories is it ok to have it in multiple categories, or should I only have the image in the most specific category?

You should use the CSV approach in this case, like we did in the Planet dataset.

1 Like

Can you remind me what the directory structure looked like for this?

The dataset isn’t available any longer, unfortunately, and I don’t have the data saved anywhere since I deleted my paperspace instance. Is that dataset somewhere I can snag so I can see how it was laid out again?

Ah, I actually still have the dataset on my aws instance. I see how it is working now. Thanks Jeremy!

Is the planet_cv notebook in a non-working order right now? I think I have everything up to date, but it is not finding the imports.

I took a step back and started a new notebook from scratch and am piecing together the elements I need. So far so good…thanks!

The notebook you want is ‘lesson2-image_models.ipynb’

1 Like