|
Ok, I've trained object classifier using randomly sampled patches. Now how can I use it for detection assuming all objects are same size but each image can contain up to a hundred of such (small) objects? Ideally I need to process whole image in a sliding window mode and assign a probability to each pixel (cause objects can be dense). It could be really hard doing it in naive way. So now I'm just replacing last full layer with convolutional layer and do exactly the same as for training but on larger images. But it is still time-consuming. What is the usual approach for the problem? Selective search or pyramid pooling layer? Thanks! |
