morten...@gmail.com
Hi
I am new to Caffe. My problem sound simple, but I haven’t been able to find an answer.
I have created a small model based on imagenet. The model has to classes (Cats and Dogs).
When I use the script (python/classify.py) with images of cats and dogs, the model is working pretty create. But when I give it an image of something completely different like a car, the result is fx cat 0.45 dog 0.55. It seems like the sum of all the classes is always 1.
I have been trying to use the script (python/detect.py), but it requires MATLAB to get the selective search working, and I don’t have MATLAB.
I would like to make a
model that outputs true or false for each class.
I have been thinking of two solutions.
1: To add a class of random images named unclassified to trigger the images that can’t be classified the other classes. I have used that method on simple feedforward NN for facial detection successfully, but I don’t know well it will work for CNN.
2: To make a simple cutoff (in python) the makes sure that the class is above 0.99 to be set true.
There are likely a better solution than these two.
Have anyone dealt with this problem (it seems where general).
I would appreciate any inputs a lot.
Thanks,
Morten