I'm using DLIB's object detection example and trying to train it using 7 images containing 14 images of a bottle. These images are primarily around 200x300 pixels, although 2 are much larger (the 1500x2000 pixel realm). The large images contain only 1 example each, and although the images are really big, the bottles themselves match roughly the same size of the bottles in the smaller training images. My sliding window is 70x240 which is about the average size of the bounding boxes I drew.
It has now been minimizing the objective function for 8+ hours on a Windows Server Machine with 384GB of RAM running Windows 8 64bit. There's no way it's supposed to take this long. It's still going--it's on iteration 125...
The documentation mentions training a face detector on the provided set of faces in "on the order of 10 seconds." Could it be because I'm running in MS Visual Studio 2012 with arguments passed into the debugger? Even when I ran the face detector example, it took a solid 30-45 minutes for training--a lot more than 10 seconds.
Has anyone has similar issues and know how to fix it?
Thanks for your help!
Did you compile in debug mode? See: http://dlib.net/faq.html#Whyisdlibslow