What are the Coco-stuff segmentation colors for all of the 182 classes?

[AugmentedRealityCat]

I have read from the issues page that the T2i Semantic Segmentation model had been trained on 182 classes defined by the Coco-Stuff model.

I have found the reference document for those 182 classes over here: https://github.com/nightrome/cocostuff/blob/master/labels.txt

And here is what @MC-E wrote about the color mapping of those 182 categories:

The segmentation map includes the labels. Here, we use the coco-stuff as the training data, with 182 classes. Each class corresponds to a pixel value (0-181) in the segmentation map.

I've tried putting the values as RGB, for example RGB(153,153,153) for Salad since it's the number associated with that class, but it doesn't work. This was saved as a PNG in 8bpc.

My question is this: Where can we find a spreadsheet or some similar document that will tell us precisely what are the corresponding RGB color values for each of the 182 classes?

And if there is no such document, where can I find the data to make one ?

[AugmentedRealityCat]

Since I haven't found any reference telling me which color goes with each of the 182 classes, I decided to build one myself. But manually because I am not a programmer and I tried to understand what the code did, but hey it's just way way out of my league.

So I go through each class name, I search for a good photo on google, I send it to the segmentation pre-processor, I load it in photoshop, I extract the color of the object I want to identify, and I copy paste its values in a spreadsheet. I'm at number 75 out of 182 - almost halfway there ! (let's be positive !).

I'll share it here, but probably only tomorrow as it will take many more hours to complete it. I don't want it to have errors, so I have to double check everything.

Hopefully this will be useful for some of you.

[AugmentedRealityCat]

Here is the complete version of the T2i Semantic Segmentation Color Reference Chart (v21). This should be the final version - it is complete and all data has been cross-checked with the official code. Get it here:
Mikubill/sd-webui-controlnet#503