4

My training set has 500 positive images and 1000 negative images. My rig's memory capacity is 7.7GB and of which 31% is normally used. I tried to allocate about 4 GB for the training but it gets error. Strangely it says failed to allocate 1569600004 bytes. enter image description here

Sgt. Pepper
  • 177
  • 11
  • You might want to check the dimension (size) of your images. I believe Haar features are extracted from the images as feature vectors. So, if your image is too big, you end up with too many feature vectors. This training program was designed for training a modified Adaboost classifier for face detection. Hence, the training images are small images of faces in the order of 100 x 100 pixels, not megapixel images. – lightalchemist Dec 23 '13 at 04:30
  • Does that mean I need to reduce the image resolution before marking/cropping the image for training? – Sgt. Pepper Dec 24 '13 at 04:14
  • What are the sizes of your training images like and what sort of objects are you trying to detect? The error says your computer is unable to allocate approximately 1.46Gb of memory. So it might just be a case that you do not have enough ram. Note that this training does take a lot of time and memory. It is hard to say unless you provide more info. – lightalchemist Dec 24 '13 at 04:31
  • Previously I tried with faces and some of them were large. Right now I am working with hand images. They are 160x120 300 dpi images. I marked them all and created 24x24 samples. This time I tried to allocate 800MB and 200 MB. Still, it can't allocate enough memory. My rig has 8GB of memory. – Sgt. Pepper Dec 24 '13 at 07:10

0 Answers0