glenn-jocher commited on
Commit
19e0208
1 Parent(s): b73c62e

Update hyp.scratch-high.yaml (#6525)

Browse files

Update `lrf: 0.1`, tested on YOLOv5x6 to 55.0 mAP@0.5:0.95, slightly higher than current.

Files changed (1) hide show
  1. data/hyps/hyp.scratch-high.yaml +1 -1
data/hyps/hyp.scratch-high.yaml CHANGED
@@ -4,7 +4,7 @@
4
  # See tutorials for hyperparameter evolution https://github.com/ultralytics/yolov5#tutorials
5
 
6
  lr0: 0.01 # initial learning rate (SGD=1E-2, Adam=1E-3)
7
- lrf: 0.2 # final OneCycleLR learning rate (lr0 * lrf)
8
  momentum: 0.937 # SGD momentum/Adam beta1
9
  weight_decay: 0.0005 # optimizer weight decay 5e-4
10
  warmup_epochs: 3.0 # warmup epochs (fractions ok)
 
4
  # See tutorials for hyperparameter evolution https://github.com/ultralytics/yolov5#tutorials
5
 
6
  lr0: 0.01 # initial learning rate (SGD=1E-2, Adam=1E-3)
7
+ lrf: 0.1 # final OneCycleLR learning rate (lr0 * lrf)
8
  momentum: 0.937 # SGD momentum/Adam beta1
9
  weight_decay: 0.0005 # optimizer weight decay 5e-4
10
  warmup_epochs: 3.0 # warmup epochs (fractions ok)