3 ms·
Thanks for the reply - sorry if it wasn't clear but by "expert" I meant an expert on e.g. early 20th century clothing or something like that who would confirm o
by SleekEagle 5y ago
Thanks for the reply - sorry if it wasn't clear but by "expert" I meant an expert on e.g. early 20th century clothing or something like that who would confirm or refute the performance of the recolorization algorithm
- PaulHoule 5y agoSure. If it's important to you that the colors come out like the original you are going to need that expert. That expert might very well tell you that the problem is well-determined in some cases (I don't think you'll ever see that rug pattern with different colors) and in other cases isn't (right now I am wearing a hoodie I bought at Target that comes in I think 7 different colors, so all an expert can tell you is that it could be one of many colors.) That's an unappreciated point in machine learning. In some cases we can classify something accurately, in some cases the answer is always going to be ambiguous. From a practical aspect some answers are critical to the success of the application, in other cases it makes no difference if item X is classified in class A or class B. Developing a training set you can drive yourself batty accurately classifying things that don't matter and still have a training set that's worthless because the important things aren't determined.
- SleekEagle 5y agoFair enough and good points. Thanks for the reply, looks like I was too reactive. Appreciate your comments!