1

I am trying to train an object detector using D-lib. I selected close to 100 images for training. I am using the Python environment. As per documentation, I used the Imglab tool to draw the bounding boxes across the images. Every image is almost 4000*3000 pixels in size. And then placed the generated XML file into my location and called the detector program. Here are my doubts and questions.

What should I use as the testing XML file while running the program? I ran without assigning any testing XML in place. However, I get the following. Training with C: 5 Training with epsilon: 0.01 Training using 8 threads. Training with sliding window 81 pixels wide by 79 pixels tall. Training on both left and right flipped versions of images. Killed What does 'killed' mean by the way?

What should I do now? Please guide!

Praveen
  • 174
  • 15
  • 4000*3000 is too large for sliding windth 81x79. Yes, it can train - but will take enormous amount of RAM. Use small images - crop them with your object + some space around them – Evgeniy Aug 19 '16 at 06:48
  • Thanks! It worked when I downsampled the images, it worked fine. Got a trained xml and svm file. So, how do I use the trained classifier in my own imageset? – Praveen Aug 22 '16 at 15:12
  • after classifier is trained - it will work with any amage size lager than 81x79 – Evgeniy Aug 22 '16 at 16:39
  • So, the xml file becomes the classifier! What is the use of the svm file? – Praveen Aug 23 '16 at 23:35
  • here is the sample http://dlib.net/train_object_detector.py.html. you can construct (load) detector from .svm file. and you can train object detector without xml – Evgeniy Aug 24 '16 at 06:30

1 Answers1

1

The testing XML file gives images and object annotations to be used to check the accuracy of your object detection. The file can be generated using imglab as with the training XML. It should contain data that is similar to, but not identical to, your training dataset.

The Killed message is due to memory issues. In Linux, it is the result of the OOM Killer. Since the images are so large, your machine is running out of memory and killing the training process.
Possible solutions:
1) Configure the OOM killer to allow dlib to use more memory.
2) Scale down the images if possible for your application.
3) Use a machine with more RAM. If you don't have one on hand, AWS offers several high-RAM EC2 options at relatively low cost.

bslawski
  • 600
  • 1
  • 9
  • 22