10月24日 炼丹

李志颖

日志记录：10月24日早8点 代号：001

:::info 数据：采用yolov5x模型，采用的是原始的超参数文件
命令行：
python train.py —weights weights/yolov5x.pt —cfg models/yolov5x.yaml —hyp data/hyps/hyp.scratch.yaml —data dataSet/clothes.yaml —batch-size 16 —epochs 300 —device 6,7 :::
lr0: 0.01
lrf: 0.2
momentum: 0.937
weight_decay: 0.0005
warmup_epochs: 3.0
warmup_momentum: 0.8
warmup_bias_lr: 0.1
box: 0.05
cls: 0.5
cls_pw: 1.0
obj: 1.0
obj_pw: 1.0
iou_t: 0.2
anchor_t: 4.0
fl_eiou_gamma: 0.0
iou_aware: 0.0
fl_gamma: 0.0
hsv_h: 0.015
hsv_s: 0.7
hsv_v: 0.4
degrees: 0.0
translate: 0.1
scale: 0.5
shear: 0.0
perspective: 0.0
flipud: 0.0
fliplr: 0.5
mosaic: 1.0
mixup: 0.0
copy_paste: 0.0

日志记录：10月24日早8点 代号：002

:::info 数据：将box从0.04改成了0.05,iou_t从0.20改成0.25
命令行：
python train.py —weights weights/yolov5s6.pt —cfg models/yolov5s6-bifpn_atten.yaml —hyp data/hyps/hyp.best2.yaml —data dataSet/clothes.yaml —batch-size 16 —epochs 300 —device 0,1,2,3 :::
# 修改点，将box从0.04改成了0.05,iou_t从0.20改成0.25
lr0: 0.01 # initial learning rate (SGD=1E-2, Adam=1E-3)
lrf: 0.2 # final OneCycleLR learning rate (lr0 * lrf)
momentum: 0.937 # SGD momentum/Adam beta1
weight_decay: 0.0005 # optimizer weight decay 5e-4
warmup_epochs: 3.0 # warmup epochs (fractions ok)
warmup_momentum: 0.8 # warmup initial momentum
warmup_bias_lr: 0.1 # warmup initial bias lr
box: 0.05 # box loss gain
cls: 0.4 # cls loss gain
cls_pw: 1.0 # cls BCELoss positive_weight
obj: 1.0 # obj loss gain (scale with pixels)
obj_pw: 1.0 # obj BCELoss positive_weight
#iou_t: 0.20 # IoU training threshold
iou_t: 0.25 # IoU training threshold
anchor_t: 4.0 # anchor-multiple threshold
# anchors: 3 # anchors per output layer (0 to ignore)
fl_eiou_gamma: 0.0 #focal eiou loss gamma
iou_aware: 0.0
fl_gamma: 0.0 # focal loss gamma (efficientDet default gamma=1.5)
hsv_h: 0.015 # image HSV-Hue augmentation (fraction)
hsv_s: 0.7 # image HSV-Saturation augmentation (fraction)
hsv_v: 0.4 # image HSV-Value augmentation (fraction)
degrees: 0.0 # image rotation (+/- deg)
translate: 0.1 # image translation (+/- fraction)
scale: 0.5 # image scale (+/- gain)
shear: 0.0 # image shear (+/- deg)
perspective: 0.0 # image perspective (+/- fraction), range 0-0.001
flipud: 0.0 # image flip up-down (probability)
fliplr: 0.5 # image flip left-right (probability)
mosaic: 1.0 # image mosaic (probability)
mixup: 0.0 # image mixup (probability)
copy_paste: 0.0 # segment copy-paste (probability)

日志记录：10月24日早8点 代号：003

:::info 数据：将iout从0.25改成了0.28
命令行：
python train.py —weights weights/yolov5s6.pt —cfg models/yolov5s6-bifpn_atten.yaml —hyp data/hyps/hyp.best3.yaml —data dataSet/clothes.yaml —batch-size 16 —epochs 300 —device 4,5 :::
# 修改点，将iout从0.25改成了0.28
lr0: 0.01 # initial learning rate (SGD=1E-2, Adam=1E-3)
lrf: 0.2 # final OneCycleLR learning rate (lr0 * lrf)
momentum: 0.937 # SGD momentum/Adam beta1
weight_decay: 0.0005 # optimizer weight decay 5e-4
warmup_epochs: 3.0 # warmup epochs (fractions ok)
warmup_momentum: 0.8 # warmup initial momentum
warmup_bias_lr: 0.1 # warmup initial bias lr
box: 0.05 # box loss gain
cls: 0.4 # cls loss gain
cls_pw: 1.0 # cls BCELoss positive_weight
obj: 1.0 # obj loss gain (scale with pixels)
obj_pw: 1.0 # obj BCELoss positive_weight
#iou_t: 0.20 # IoU training threshold
#iou_t: 0.25 # IoU training threshold
iou_t: 0.28 # IoU training threshold
anchor_t: 4.0 # anchor-multiple threshold
# anchors: 3 # anchors per output layer (0 to ignore)
fl_eiou_gamma: 0.0 #focal eiou loss gamma
iou_aware: 0.0
fl_gamma: 0.0 # focal loss gamma (efficientDet default gamma=1.5)
hsv_h: 0.015 # image HSV-Hue augmentation (fraction)
hsv_s: 0.7 # image HSV-Saturation augmentation (fraction)
hsv_v: 0.4 # image HSV-Value augmentation (fraction)
degrees: 0.0 # image rotation (+/- deg)
translate: 0.1 # image translation (+/- fraction)
scale: 0.5 # image scale (+/- gain)
shear: 0.0 # image shear (+/- deg)
perspective: 0.0 # image perspective (+/- fraction), range 0-0.001
flipud: 0.0 # image flip up-down (probability)
fliplr: 0.5 # image flip left-right (probability)
mosaic: 1.0 # image mosaic (probability)
mixup: 0.0 # image mixup (probability)
copy_paste: 0.0 # segment copy-paste (probability)

日志记录：10月24日早8点 代号：004

:::info 数据：修改点，将iout从0.28改成了0.30
命令行：
python train.py —weights weights/yolov5s6.pt —cfg models/yolov5s6-bifpn_atten.yaml —hyp data/hyps/hyp.best4.yaml —data dataSet/clothes.yaml —batch-size 16 —epochs 300 —device 0,1,2,3 :::
# 修改点，将iout从0.28改成了0.30
lr0: 0.01 # initial learning rate (SGD=1E-2, Adam=1E-3)
lrf: 0.2 # final OneCycleLR learning rate (lr0 * lrf)
momentum: 0.937 # SGD momentum/Adam beta1
weight_decay: 0.0005 # optimizer weight decay 5e-4
warmup_epochs: 3.0 # warmup epochs (fractions ok)
warmup_momentum: 0.8 # warmup initial momentum
warmup_bias_lr: 0.1 # warmup initial bias lr
box: 0.05 # box loss gain
cls: 0.4 # cls loss gain
cls_pw: 1.0 # cls BCELoss positive_weight
obj: 1.0 # obj loss gain (scale with pixels)
obj_pw: 1.0 # obj BCELoss positive_weight
#iou_t: 0.20 # IoU training threshold
#iou_t: 0.25 # IoU training threshold
#iou_t: 0.28 # IoU training threshold
iou_t: 0.30 # IoU training threshold
anchor_t: 4.0 # anchor-multiple threshold
# anchors: 3 # anchors per output layer (0 to ignore)
fl_eiou_gamma: 0.0 #focal eiou loss gamma
iou_aware: 0.0
fl_gamma: 0.0 # focal loss gamma (efficientDet default gamma=1.5)
hsv_h: 0.015 # image HSV-Hue augmentation (fraction)
hsv_s: 0.7 # image HSV-Saturation augmentation (fraction)
hsv_v: 0.4 # image HSV-Value augmentation (fraction)
degrees: 0.0 # image rotation (+/- deg)
translate: 0.1 # image translation (+/- fraction)
scale: 0.5 # image scale (+/- gain)
shear: 0.0 # image shear (+/- deg)
perspective: 0.0 # image perspective (+/- fraction), range 0-0.001
flipud: 0.0 # image flip up-down (probability)
fliplr: 0.5 # image flip left-right (probability)
mosaic: 1.0 # image mosaic (probability)
mixup: 0.0 # image mixup (probability)
copy_paste: 0.0 # segment copy-paste (probability)

日志记录：10月24日早8点 代号：005

:::info 数据：修改点，将cls从0.4改成了0.35
命令行：
python train.py —weights weights/yolov5s6.pt —cfg models/yolov5s6-bifpn_atten.yaml —hyp data/hyps/hyp.best5.yaml —data dataSet/clothes.yaml —batch-size 16 —epochs 300 —device 0,1,2,3 ::: # 修改点，将cls从0.4改成了0.35
lr0: 0.01 # initial learning rate (SGD=1E-2, Adam=1E-3)
lrf: 0.2 # final OneCycleLR learning rate (lr0 * lrf)
momentum: 0.937 # SGD momentum/Adam beta1
weight_decay: 0.0005 # optimizer weight decay 5e-4
warmup_epochs: 3.0 # warmup epochs (fractions ok)
warmup_momentum: 0.8 # warmup initial momentum
warmup_bias_lr: 0.1 # warmup initial bias lr
box: 0.05 # box loss gain
#cls: 0.4 # cls loss gain
cls: 0.35 # cls loss gain
cls_pw: 1.0 # cls BCELoss positive_weight
obj: 1.0 # obj loss gain (scale with pixels)
obj_pw: 1.0 # obj BCELoss positive_weight
#iou_t: 0.20 # IoU training threshold
#iou_t: 0.25 # IoU training threshold
iou_t: 0.28 # IoU training threshold
anchor_t: 4.0 # anchor-multiple threshold
# anchors: 3 # anchors per output layer (0 to ignore)
fl_eiou_gamma: 0.0 #focal eiou loss gamma
iou_aware: 0.0
fl_gamma: 0.0 # focal loss gamma (efficientDet default gamma=1.5)
hsv_h: 0.015 # image HSV-Hue augmentation (fraction)
hsv_s: 0.7 # image HSV-Saturation augmentation (fraction)
hsv_v: 0.4 # image HSV-Value augmentation (fraction)
degrees: 0.0 # image rotation (+/- deg)
translate: 0.1 # image translation (+/- fraction)
scale: 0.5 # image scale (+/- gain)
shear: 0.0 # image shear (+/- deg)
perspective: 0.0 # image perspective (+/- fraction), range 0-0.001
flipud: 0.0 # image flip up-down (probability)
fliplr: 0.5 # image flip left-right (probability)
mosaic: 1.0 # image mosaic (probability)
mixup: 0.0 # image mixup (probability)
copy_paste: 0.0 # segment copy-paste (probability)

日志记录：10月24日早8点 代号：006

:::info 数据：采用最好的best3超参数文件，epoch改为了400
命令行：
python train.py —weights weights/yolov5s6.pt —cfg models/yolov5s6-bifpn_atten.yaml —hyp data/hyps/hyp.best3.yaml —data dataSet/clothes.yaml —batch-size 16 —epochs 400 —device 4,5,6,7 :::
# 修改点，将iout从0.25改成了0.28
lr0: 0.01 # initial learning rate (SGD=1E-2, Adam=1E-3)
lrf: 0.2 # final OneCycleLR learning rate (lr0 * lrf)
momentum: 0.937 # SGD momentum/Adam beta1
weight_decay: 0.0005 # optimizer weight decay 5e-4
warmup_epochs: 3.0 # warmup epochs (fractions ok)
warmup_momentum: 0.8 # warmup initial momentum
warmup_bias_lr: 0.1 # warmup initial bias lr
box: 0.05 # box loss gain
cls: 0.4 # cls loss gain
cls_pw: 1.0 # cls BCELoss positive_weight
obj: 1.0 # obj loss gain (scale with pixels)
obj_pw: 1.0 # obj BCELoss positive_weight
#iou_t: 0.20 # IoU training threshold
#iou_t: 0.25 # IoU training threshold
iou_t: 0.28 # IoU training threshold
anchor_t: 4.0 # anchor-multiple threshold
# anchors: 3 # anchors per output layer (0 to ignore)
fl_eiou_gamma: 0.0 #focal eiou loss gamma
iou_aware: 0.0
fl_gamma: 0.0 # focal loss gamma (efficientDet default gamma=1.5)
hsv_h: 0.015 # image HSV-Hue augmentation (fraction)
hsv_s: 0.7 # image HSV-Saturation augmentation (fraction)
hsv_v: 0.4 # image HSV-Value augmentation (fraction)
degrees: 0.0 # image rotation (+/- deg)
translate: 0.1 # image translation (+/- fraction)
scale: 0.5 # image scale (+/- gain)
shear: 0.0 # image shear (+/- deg)
perspective: 0.0 # image perspective (+/- fraction), range 0-0.001
flipud: 0.0 # image flip up-down (probability)
fliplr: 0.5 # image flip left-right (probability)
mosaic: 1.0 # image mosaic (probability)
mixup: 0.0 # image mixup (probability)
copy_paste: 0.0 # segment copy-paste (probability)

日志记录：10月24日早8点 代号：007

:::info 数据：将cls从0.35改成了0.33
命令行：
python train.py —weights weights/yolov5s6.pt —cfg models/yolov5s6-bifpn_atten.yaml —hyp data/hyps/hyp.best6.yaml —data dataSet/clothes.yaml —batch-size 16 —epochs 400 —device 0,1,2,3 :::
# 修改点，将cls从0.35改成了0.33
lr0: 0.01 # initial learning rate (SGD=1E-2, Adam=1E-3)
lrf: 0.2 # final OneCycleLR learning rate (lr0 * lrf)
momentum: 0.937 # SGD momentum/Adam beta1
weight_decay: 0.0005 # optimizer weight decay 5e-4
warmup_epochs: 3.0 # warmup epochs (fractions ok)
warmup_momentum: 0.8 # warmup initial momentum
warmup_bias_lr: 0.1 # warmup initial bias lr
box: 0.05 # box loss gain
#cls: 0.4 # cls loss gain
#cls: 0.35 # cls loss gain
cls: 0.33 # cls loss gain
cls_pw: 1.0 # cls BCELoss positive_weight
obj: 1.0 # obj loss gain (scale with pixels)
obj_pw: 1.0 # obj BCELoss positive_weight
#iou_t: 0.20 # IoU training threshold
#iou_t: 0.25 # IoU training threshold
iou_t: 0.28 # IoU training threshold
anchor_t: 4.0 # anchor-multiple threshold
# anchors: 3 # anchors per output layer (0 to ignore)
fl_eiou_gamma: 0.0 #focal eiou loss gamma
iou_aware: 0.0
fl_gamma: 0.0 # focal loss gamma (efficientDet default gamma=1.5)
hsv_h: 0.015 # image HSV-Hue augmentation (fraction)
hsv_s: 0.7 # image HSV-Saturation augmentation (fraction)
hsv_v: 0.4 # image HSV-Value augmentation (fraction)
degrees: 0.0 # image rotation (+/- deg)
translate: 0.1 # image translation (+/- fraction)
scale: 0.5 # image scale (+/- gain)
shear: 0.0 # image shear (+/- deg)
perspective: 0.0 # image perspective (+/- fraction), range 0-0.001
flipud: 0.0 # image flip up-down (probability)
fliplr: 0.5 # image flip left-right (probability)
mosaic: 1.0 # image mosaic (probability)
mixup: 0.0 # image mixup (probability)
copy_paste: 0.0 # segment copy-paste (probability)

日志记录：10月24日早8点 代号：008

:::info 数据：修改点，将cls从0.33改成了0.30
命令行：
python train.py —weights weights/yolov5s6.pt —cfg models/yolov5s6-bifpn_atten.yaml —hyp data/hyps/hyp.best7.yaml —data dataSet/clothes.yaml —batch-size 16 —epochs 400 —device 4,5,6,7 :::
# 修改点，将cls从0.33改成了0.30
lr0: 0.01 # initial learning rate (SGD=1E-2, Adam=1E-3)
lrf: 0.2 # final OneCycleLR learning rate (lr0 * lrf)
momentum: 0.937 # SGD momentum/Adam beta1
weight_decay: 0.0005 # optimizer weight decay 5e-4
warmup_epochs: 3.0 # warmup epochs (fractions ok)
warmup_momentum: 0.8 # warmup initial momentum
warmup_bias_lr: 0.1 # warmup initial bias lr
box: 0.05 # box loss gain
#cls: 0.4 # cls loss gain
#cls: 0.35 # cls loss gain
#cls: 0.33 # cls loss gain
cls: 0.30 # cls loss gain
cls_pw: 1.0 # cls BCELoss positive_weight
obj: 1.0 # obj loss gain (scale with pixels)
obj_pw: 1.0 # obj BCELoss positive_weight
#iou_t: 0.20 # IoU training threshold
#iou_t: 0.25 # IoU training threshold
iou_t: 0.28 # IoU training threshold
anchor_t: 4.0 # anchor-multiple threshold
# anchors: 3 # anchors per output layer (0 to ignore)
fl_eiou_gamma: 0.0 #focal eiou loss gamma
iou_aware: 0.0
fl_gamma: 0.0 # focal loss gamma (efficientDet default gamma=1.5)
hsv_h: 0.015 # image HSV-Hue augmentation (fraction)
hsv_s: 0.7 # image HSV-Saturation augmentation (fraction)
hsv_v: 0.4 # image HSV-Value augmentation (fraction)
degrees: 0.0 # image rotation (+/- deg)
translate: 0.1 # image translation (+/- fraction)
scale: 0.5 # image scale (+/- gain)
shear: 0.0 # image shear (+/- deg)
perspective: 0.0 # image perspective (+/- fraction), range 0-0.001
flipud: 0.0 # image flip up-down (probability)
fliplr: 0.5 # image flip left-right (probability)
mosaic: 1.0 # image mosaic (probability)
mixup: 0.0 # image mixup (probability)
copy_paste: 0.0 # segment copy-paste (probability)

日志记录：10月24日早8点 代号：009

:::info 数据：修改点，将cls从0.4改成了0.45
命令行：
python train.py —weights weights/yolov5s6.pt —cfg models/yolov5s6-bifpn_atten.yaml —hyp data/hyps/hyp.best8.yaml —data dataSet/clothes.yaml —batch-size 16 —epochs 400 —device 0,1,2,3 :::
# 修改点，将cls从0.4改成了0.45
lr0: 0.01 # initial learning rate (SGD=1E-2, Adam=1E-3)
lrf: 0.2 # final OneCycleLR learning rate (lr0 * lrf)
momentum: 0.937 # SGD momentum/Adam beta1
weight_decay: 0.0005 # optimizer weight decay 5e-4
warmup_epochs: 3.0 # warmup epochs (fractions ok)
warmup_momentum: 0.8 # warmup initial momentum
warmup_bias_lr: 0.1 # warmup initial bias lr
box: 0.05 # box loss gain
#cls: 0.4 # cls loss gain
cls: 0.45 # cls loss gain
cls_pw: 1.0 # cls BCELoss positive_weight
obj: 1.0 # obj loss gain (scale with pixels)
obj_pw: 1.0 # obj BCELoss positive_weight
#iou_t: 0.20 # IoU training threshold
#iou_t: 0.25 # IoU training threshold
iou_t: 0.28 # IoU training threshold
anchor_t: 4.0 # anchor-multiple threshold
# anchors: 3 # anchors per output layer (0 to ignore)
fl_eiou_gamma: 0.0 #focal eiou loss gamma
iou_aware: 0.0
fl_gamma: 0.0 # focal loss gamma (efficientDet default gamma=1.5)
hsv_h: 0.015 # image HSV-Hue augmentation (fraction)
hsv_s: 0.7 # image HSV-Saturation augmentation (fraction)
hsv_v: 0.4 # image HSV-Value augmentation (fraction)
degrees: 0.0 # image rotation (+/- deg)
translate: 0.1 # image translation (+/- fraction)
scale: 0.5 # image scale (+/- gain)
shear: 0.0 # image shear (+/- deg)
perspective: 0.0 # image perspective (+/- fraction), range 0-0.001
flipud: 0.0 # image flip up-down (probability)
fliplr: 0.5 # image flip left-right (probability)
mosaic: 1.0 # image mosaic (probability)
mixup: 0.0 # image mixup (probability)
copy_paste: 0.0 # segment copy-paste (probability)

日志记录：10月24日早8点 代号：010

:::info 数据：iou_t从0.28调整到0.32
命令行：
python train.py —weights weights/yolov5s6.pt —cfg models/yolov5s6-bifpn_atten.yaml —hyp data/hyps/hyp.best9.yaml —data dataSet/clothes.yaml —batch-size 16 —epochs 400 —device 0,1,2,3 :::
# iou_t从0.28调整到0.32
lr0: 0.01 # initial learning rate (SGD=1E-2, Adam=1E-3)
lrf: 0.02 # final OneCycleLR learning rate (lr0 * lrf)
momentum: 0.9 # SGD momentum/Adam beta1
weight_decay: 0.0005 # optimizer weight decay 5e-4
warmup_epochs: 3.0 # warmup epochs (fractions ok)
warmup_momentum: 0.8 # warmup initial momentum
warmup_bias_lr: 0.1 # warmup initial bias lr
box: 0.05 # box loss gain
cls: 0.50 # cls loss gain
cls_pw: 1.0 # cls BCELoss positive_weight
obj: 0.50 # obj loss gain (scale with pixels)
obj_pw: 1.0 # obj BCELoss positive_weight
iou_t: 0.32 # IoU training threshold
anchor_t: 4.0 # anchor-multiple threshold
# anchors: 3 # anchors per output layer (0 to ignore)
fl_eiou_gamma: 0.0 #focal eiou loss gamma
iou_aware: 0.0
fl_gamma: 0.0 # focal loss gamma (efficientDet default gamma=1.5)
hsv_h: 0.015 # image HSV-Hue augmentation (fraction)
hsv_s: 0.7 # image HSV-Saturation augmentation (fraction)
hsv_v: 0.4 # image HSV-Value augmentation (fraction)
degrees: 0.0 # image rotation (+/- deg)
translate: 0.1 # image translation (+/- fraction)
scale: 0.5 # image scale (+/- gain)
shear: 0.0 # image shear (+/- deg)
perspective: 0.0 # image perspective (+/- fraction), range 0-0.001
flipud: 0.0 # image flip up-down (probability)
fliplr: 0.5 # image flip left-right (probability)
mosaic: 1.0 # image mosaic (probability)
mixup: 0.0 # image mixup (probability)
copy_paste: 0.0 # segment copy-paste (probability)