search

AlexeyAB / darknet

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )
http://pjreddie.com/darknet/
Other
21.65k stars 7.96k forks source link

If you don’t want to use mosaic data augmentation, is there any way to make up for it? #7073

Open joe660 opened 3 years ago

joe660 commented 3 years ago

I only have one category, and I need to learn from false positive samples. I cannot use it with mosaic. Does turning off mosaic have a big impact? Are there any suggested alternatives or remedies? Thank you for your reply

stephanecharette commented 3 years ago

Does turning off mosaic have a big impact?

Turning it off should be fine. Mosaic is already off by default in many configurations, e.g., yolov4-tiny, yolov4-tiny-3l.

Are there any suggested alternatives or remedies?

No, you can simply turn it off with mosaic=0.

joe660 commented 3 years ago

No, you can simply turn it off with mosaic=0.

Thank you for your prompt reply. But without mosaic, the convergence is a bit slow from the training point of view. I don't know which is better for the final result or after mosaic is added? I don’t know it can be alleviated by modifying some parameters