coreyabshire/color-names

Convex Hull based fuzzy logic classifier and associated utilities for the color naming task.

color-names's Issues

model not currently implemented as per paper

The current implementation, though working and having decent results, isn't quite as it was originally envisioned in the paper. To match what is in the paper, the beta parameters for the sigmoid should be locked between two hulls, not adjusted as they are currently. This locked value should be fit using some sort of model, based on the ratings given.

select a subset of the colors to use

We need to choose a smaller subset of the PANTONE color patches to use in the actual experiment. We think it should be based on distance from the convex hulls formed by the colors from the old data.

only full agreement

The convex hull classifier should only form the hull based on patches for which there was consensus among all observers. This is needed for the color subset selection which is needed for the experiment.

need line num in html diagnostic

implement referenced (Benavente, et. al.) model

It would be helpful to be able to check the results of the model we referenced by Benavente, et. al.

selection could exclude more on outer edge

The selection logic will select some colors that are more obvious than we would like, because it only looks at distance to the hull in which it resides. This will include colors that are on the "back" side of the hull, away from any other hulls. It would be helpful in reducing the number of "obvious" colors if we could adjust the selection formula to exclude those colors in some way, such as by looking at distance 0 and distance 1 in those cases.

selection numbers adjustment 1

After the initial review of the selection results, we feel we should base the selection on 3 Delta-E from the hull for points in hull for non-neutral colors, and 1 Delta-E for points inside the neutral hull. We should continue to include the other two classes: points outside a hull or points in more than one hull. The output should include a new column that says explicitly whether the point is considered "in selection" or not that the experiment team can use for the actual experiment.

additional test model

We need several additional test models, similar to the random and gradient ones we already have, so we can validate the model is working as expected.