search

AlexeyAB / darknet

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )
http://pjreddie.com/darknet/
Other
21.69k stars 7.96k forks source link

number of images produced by data augmentation #3586

Open jasleen137 opened 5 years ago

jasleen137 commented 5 years ago

Hi @AlexeyAB I wanted to know how many images are produced by auto-data augmentation in yolo-cfg

LukeAI commented 5 years ago

I believe that every image the network every sees is essentially unique, generated with randomised parameters (within a fairly well defined param space)

AlexeyAB commented 5 years ago

@LukeAI Yes.

BERENZ commented 5 years ago

@AlexeyAB could you please provide information what types of augmentation are applied during the training procedure? For instance, in xview-yolov3 they apply the following tasks

Augmentation Description
Translation +/- 1% (vertical and horizontal)
Rotation +/- 20 degrees
Shear +/- 3 degrees (vertical and horizontal)
Scale +/- 30%
Reflection 50% probability (vertical and horizontal)
HSV Saturation +/- 50%
HSV Intensity +/- 50%

I would like understand and be 100% sure what is done during this procedure.