LARC-CMU-SMU/FoodSeg103-Benchmark-v1

Categories and Mask Colors

Opened this issue · 5 comments

May I please ask where I can find relationship between the categories and the mask colors? I don't find any keywords of category in the code. I would greatly appreciate it if you could provide me with information.

It seems they are all loaded from the checkpoints you're using if you check init_segmentor. What is each class? I don't know at all. However I can tell you each position in checkpoint['meta']['CLASSES'] has a corresponding checkpoint['meta']['PALETTE']. If you print those you can see they are int list for classes and a list of lists of 3 ints for Palette which means basically one color for each class which I guess you already knew. However now at least we know what class number corresponds to each color haha.

Now I'll try to find what is each class wish me luck 😄

So I finally got it you see inside the Food103 dataset you'll find category_id.txt which shows all tags used for the checkpoints provided. From what I have seen all checkpoints have been trained for those 103 tags so now you know what tag, number and color corresponds for each class.

Hi @WilsonSmithRobles,

I had a quick follow-up to your previous post,
Where can I find the correspondence between the color palette and the 103 classes in the original dataset (ann_dir folder)?

Thanks!

@SamyakR99 I guess it uses a positional reference being color[0] = annotation[0]. I don't know for sure though.

@WilsonSmithRobles

I think that should be it. Thanks for the help!