Reputation: 558
I am attempting to use the haar cascade trainer GUI to train a model to detect custom images.
I keep getting the below error whenever my positive images are more than 200. I have 1256 images in my negative(n) folder and 500+ in my positive(p) folder. If I train it with less than or 200 images it will not break but it's not detecting the object accurately when tested.
Upvotes: 0
Views: 1840
Reputation: 1
This maybe to late, however just fixed similar issue and found solution somewhere else. What you need to do when having larger positive images is lower the percentage used with in the Train window. Lowering it to 90% fixed issue.
Upvotes: 0