I have something like 15000 "red", 1300 "yellow", 350 "Blue" and 200 "green" My plan was to get rid of maybe 14000 "red", 300 "yellow" and roughly triplicate "Blue" and quintiply "green" to have about the same number of samples for each coloe (well, actually I have some samples I'm not using, I'd merge those first and then do that kind of adjustments).
I was actually thinking of something similar to what you mention though, using the neural network as a "yes/no" classifier, and using several; instead of having a single one give different colors.
And in fact I used the colors in quotations because it's not red/green/blue/yellow; but "License plate background", "license plate foreground", "license plate back or foreground" and "non-license plate color" That last group is very broad, but oddly the license plate foreground is the one with the 15K samples; I also need to feed this data into my 3D visualization program to verify if the data is correct.
Comments 3
(The comment has been removed)
My plan was to get rid of maybe 14000 "red", 300 "yellow" and roughly triplicate "Blue" and quintiply "green" to have about the same number of samples for each coloe (well, actually I have some samples I'm not using, I'd merge those first and then do that kind of adjustments).
I was actually thinking of something similar to what you mention though, using the neural network as a "yes/no" classifier, and using several; instead of having a single one give different colors.
And in fact I used the colors in quotations because it's not red/green/blue/yellow; but "License plate background", "license plate foreground", "license plate back or foreground" and "non-license plate color"
That last group is very broad, but oddly the license plate foreground is the one with the 15K samples; I also need to feed this data into my 3D visualization program to verify if the data is correct.
Reply
Reply
Well, your next assignment, Mr. Phelps, will be programming Kingdom Hearts III, so enjoy the easy ride now... ;D
Reply
Leave a comment