yeast_fusion_segmenter/sample_hyperparameters.yaml at main · DessimozLab/yeast_fusion_segmenter · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
# YOLO training hyperparameters

# Learning rate settings
lr0: 0.001              # initial learning rate
lrf: 0.0001             # final learning rate (lr0 * lrf)
momentum: 0.5           # SGD momentum/Adam beta1
weight_decay: 0.0001    # optimizer weight decay

# Warmup settings
warmup_epochs: 3.0      # warmup epochs
warmup_momentum: 0.8    # warmup initial momentum
warmup_bias_lr: 0.01    # warmup initial bias lr

# Loss function weights
box: 10                 # box loss gain
cls: 5                  # cls loss gain
dfl: 0.5                # dfl loss gain

# Training settings
label_smoothing: 0.0    # label smoothing
nbs: 64                 # nominal batch size

# Augmentation settings
hsv_h: 0.01             # image HSV-Hue augmentation
hsv_s: 0.01             # image HSV-Saturation augmentation
hsv_v: 0.01             # image HSV-Value augmentation
degrees: 180.0          # image rotation (+/- deg)
translate: 0.1          # image translation (+/- fraction)
scale: 0.1              # image scale (+/- gain)
shear: 0.1              # image shear (+/- deg)
perspective: 0.0        # image perspective (+/- fraction)
flipud: 0.5             # image flip up-down (probability)
fliplr: 0.5             # image flip left-right (probability)
mosaic: 0.2             # image mosaic (probability)
mixup: 0.0              # image mixup (probability)