10月5日 炼丹

罗宇鑫

日志记录： 10月5日上午 8:00 代号：001

:::info 数据：
cls_pw 调整到0.3
命令行：
python train.py —weights weights/yolov5s6.pt —cfg models/yolov5s6-bifpn_atten.yaml —data dataSet/clothes.yaml —hyp data/hyps/hyp.clothe023.yaml —batch-size 24 —epochs 70 —device 6 :::
lr0: 0.01 # initial learning rate (SGD=1E-2, Adam=1E-3)
lrf: 0.05 # final OneCycleLR learning rate (lr0 * lrf)
momentum: 0.900 # 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: 0.055 # warmup initial bias lr
box: 0.05 # box loss gain
#cls: 0.1 # cls loss gain
cls: 0.25 # cls loss gain
#cls_pw: 1.0 # cls BCELoss positive_weight
cls_pw: 0.3 # cls BCELoss positive_weight
obj: 0.5 # obj loss gain (scale with pixels)
obj_pw: 1.0 # obj BCELoss positive_weight
iou_t: 0.30 # IoU training threshold
#anchor_t: 4 # anchor-multiple threshold
anchor_t: 2.5 # 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5日上午 8:00 代号：002

:::info 数据：
cls_pw 调整到0.7
命令行：
python train.py —weights weights/yolov5s6.pt —cfg models/yolov5s6-bifpn_atten.yaml —data dataSet/clothes.yaml —hyp data/hyps/hyp.clothe022.yaml —batch-size 24 —epochs 70 —device 4 :::
lr0: 0.01 # initial learning rate (SGD=1E-2, Adam=1E-3)
lrf: 0.05 # final OneCycleLR learning rate (lr0 * lrf)
momentum: 0.900 # 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: 0.055 # warmup initial bias lr
box: 0.05 # box loss gain
#cls: 0.1 # cls loss gain
cls: 0.25 # cls loss gain
cls_pw: 1.0 # cls BCELoss positive_weight
obj: 0.5 # obj loss gain (scale with pixels)
#obj_pw: 1.0 # obj BCELoss positive_weight
obj_pw: 0.7 # obj BCELoss positive_weight
iou_t: 0.30 # IoU training threshold
#anchor_t: 4 # anchor-multiple threshold
anchor_t: 2.5 # 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5日上午 8:00 代号：003

:::info 数据：
cls_pw 调整到0.5
命令行：
python train.py —weights weights/yolov5s6.pt —cfg models/yolov5s6-bifpn_atten.yaml —data dataSet/clothes.yaml —hyp data/hyps/hyp.clothe021.yaml —batch-size 24 —epochs 70 —device 3 :::
lr0: 0.01 # initial learning rate (SGD=1E-2, Adam=1E-3)
lrf: 0.05 # final OneCycleLR learning rate (lr0 * lrf)
momentum: 0.900 # 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: 0.055 # warmup initial bias lr
box: 0.05 # box loss gain
#cls: 0.1 # cls loss gain
cls: 0.25 # cls loss gain
#cls_pw: 1.0 # cls BCELoss positive_weight
cls_pw: 0.5 # cls BCELoss positive_weight
obj: 0.5 # obj loss gain (scale with pixels)
obj_pw: 1.0 # obj BCELoss positive_weight
iou_t: 0.30 # IoU training threshold
#anchor_t: 4 # anchor-multiple threshold
anchor_t: 2.5 # 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5日上午 8:00 代号：004

:::info 数据：
cls_pw 初始值1
命令行：
python train.py —weights weights/yolov5s6.pt —cfg models/yolov5s6-bifpn_atten.yaml —data dataSet/clothes.yaml —hyp data/hyps/hyp.clothe020.yaml —batch-size 24 —epochs 70 —device 0 ::: �
lr0: 0.01 # initial learning rate (SGD=1E-2, Adam=1E-3)
lrf: 0.05 # final OneCycleLR learning rate (lr0 lrf)
momentum: 0.900 # 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: 0.055 # warmup initial bias lr
box: 0.05 # box loss gain
#cls: 0.1 # cls loss gain
cls: 0.25 # cls loss gain
cls_pw: 1.0 # cls BCELoss positive_weight
obj: 0.5 # obj loss gain (scale with pixels)
obj_pw: 1.0 # obj BCELoss positive_weight
iou_t: 0.30 # IoU training threshold
anchor_t: 2.5 # 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5日下午 20:20 代号：005

:::info 数据：
cls_pw 调整到0.7
命令行：
python train.py —weights weights/yolov5s6.pt —cfg models/yolov5s6-bifpn_atten.yaml —data dataSet/clothes.yaml —hyp data/hyps/hyp.better003.yaml —batch-size 24 —epochs 300 —device 3 :::

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: 0.5 # cls loss gain
#cls_pw: 1.0 # cls BCELoss positive_weight
cls_pw: 0.7 # 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
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5日下午 20:20 代号：006

:::info 数据：
cls_pw 调整到0.5
命令行：
python train.py —weights weights/yolov5s6.pt —cfg models/yolov5s6-bifpn_atten.yaml —data dataSet/clothes.yaml —hyp data/hyps/hyp.better002.yaml —batch-size 24 —epochs 300 —device 4 :::
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: 0.5 # cls loss gain
#cls_pw: 1.0 # cls BCELoss positive_weight
cls_pw: 0.5 # 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
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)
�

李志颖

:::tips 经过不断的调整IOU的阈值，0.2为最佳阈值点 :::

日志记录： 10月5日上午 9:30 代号：001

:::info 数据：使用的是原始的hyp.scratch.yaml文件，未做参数调整
命令行：python train.py —weights weights/yolov5s6.pt —cfg models/yolov5s6-bifpn_atten.yaml —hyp data/hyps/hyp.scratch.yaml —data dataSet/clothes.yaml —batch-size 16 —epochs 300 —device 2,4 :::
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: 0.5 # 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
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5日上午 12:30 代号：002

:::info 数据：iout从0.2调整到了0.25
命令行：python train.py —weights weights/yolov5s6.pt —cfg models/yolov5s6-bifpn_atten.yaml —hyp data/hyps/hyp.scratch01.yaml —data dataSet/clothes.yaml —batch-size 24 —epochs 300 —device 0,2,4 :::
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: 0.5 # cls loss gain
cls_pw: 1.0 # cls BCELoss positiveweight
_obj: 1.0 # obj loss gain (scale with pixels)
obj_pw: 1.0 # obj BCELoss positiveweight
#iou_t: 0.2 # IoU training threshold
_iou_t: 0.25 # IoU training thresholdanchort: 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5日下午 19:30 代号：003

:::info 数据：iout从0.25调整到了0.18
命令行：python train.py —weights weights/yolov5s6.pt —cfg models/yolov5s6-bifpn_atten.yaml —hyp data/hyps/hyp.scratch02.yaml —data dataSet/clothes.yaml —batch-size 24 —epochs 300 —device 2,4 :::
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: 0.5 # cls loss gain
cls_pw: 1.0 # cls BCELoss positiveweight
_obj: 1.0 # obj loss gain (scale with pixels)
obj_pw: 1.0 # obj BCELoss positiveweight
#iou_t: 0.2 # IoU training threshold
#iou_t: 0.25 # IoU training threshold
_iou_t: 0.18 # IoU training thresholdanchort: 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5日下午 23:50 代号：004

:::info 数据：iout从0.18调整到了0.22
命令行：python train.py —weights weights/yolov5s6.pt —cfg models/yolov5s6-bifpn_atten.yaml —hyp data/hyps/hyp.scratch03.yaml —data dataSet/clothes.yaml —batch-size 24 —epochs 300 —device 2,4 :::
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: 0.5 # cls loss gain
cls_pw: 1.0 # cls BCELoss positiveweight
_obj: 1.0 # obj loss gain (scale with pixels)
obj_pw: 1.0 # obj BCELoss positiveweight
#iou_t: 0.2 # IoU training threshold
#iou_t: 0.25 # IoU training threshold
_iou_t: 0.22 # IoU training thresholdanchort: 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)_

陈唯彬

日志记录： 10月5日下午 23:00 代号：001

:::info 数据：
根据某博主模型参考，尝试修改数据增强参数—— mixup 由0调整到0.243（增强了训练样本之间的线性表达）
命令行：
python train.py —weights weights/yolov5s6.pt —cfg models/yolov5s6-bifpnatten.yaml —hyp data/hyps/hyp.scratch.yaml —data dataSet/clothes.yaml —batch-size 16 —epochs 300 —device 2,4 :::
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: 0.5 # 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
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: 0.243 # image mixup (probability)
copy_paste: 0.0 # segment copy-paste (probability)

日志记录： 10月5日下午 23:00 代号：002

:::info 数据：
degrees~~ 由0调整到0.373（~~图片旋转角度）
命令行：
python train.py —weights weights/yolov5s6.pt —cfg models/yolov5s6-bifpn_atten.yaml —hyp data/hyps/hyp.scratch_t1.yaml —data dataSet/clothes.yaml —batch-size 16 —epochs 300 —device 2,4 :::
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: 0.5 # 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
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: 0.373 # 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: 0 # image mosaic (probability)
mixup: 0.243 # image mixup (probability)
copy_paste: 0.0 # segment copy-paste (probability)