Our team is looking to detect the FRC logo on the totes and I’ve run into more than a few problems with the training process.
I’ve done some work with pre-made datasets, but making my own has proved to be a challenge. Has any team had any luck with cascade classifier training or is there a better way to find the FIRST logo?
If anyone has had luck, how many positive samples did you use? How many negatives? How many stages? There are so many parameters to fiddle with and it would be nice to know if someone has been successful before.
When you get a chance, could you please benchmark the algorithm for how long it take to identify the object? There isn’t good literature on low sample sized classifying algorithms.
Thank you so much. So it isn’t too slow. When you have a lot of positive and negative image samples and what not, you sometimes see performances of about 5 fps on good machines.