罗宇鑫

:::tips cls 比较好的值为0.05 和 0.3
中间值0.2 反而不太好

  • 0.05 代号007
  • 0.3 代号003

0.05 的 metrics/precision 没0.3的好
但是其他更加顺滑 :::

日志记录：10月3日上午11:00 代号：001

:::info 数据：
warmup_bias_lr从 0.055 下降到 0.02
命令行：
python train.py —weights weights/yolov5s.pt —cfg models/yolov5s6-bifpn_atten.yaml —data dataSet/clothes.yaml —hyp data/hyps/hyp.clothe005.yaml —batch-size 36 —epochs 70 —device 3,4 ::: image.png
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
warmup_bias_lr: 0.02 # 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: 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: 4.0 # 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3日下午12:45 代号：002

:::info 数据：
warmup_bias_lr从 0.02 恢复到 0.06
命令行：
python train.py —weights weights/yolov5s.pt —cfg models/yolov5s6-bifpn_atten.yaml —data dataSet/clothes.yaml —hyp data/hyps/hyp.clothe007.yaml —batch-size 36 —epochs 70 —device 4,5 ::: image.png
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: 0.02 # warmup initial bias lr
warmup_bias_lr: 0.06 # 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: 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: 4.0 # 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3日下午 21:00 代号：003

:::info 数据：
cls 从 0.5 下降到 0.3
命令行：
python train.py —weights weights/yolov5s.pt —cfg models/yolov5s6-bifpn_atten.yaml —data dataSet/clothes.yaml —hyp data/hyps/hyp.clothe008.yaml —batch-size 36 —epochs 70 —device 4 ::: image.png
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: 0.5 # cls loss gain
cls: 0.3 # 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: 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3日下午 21:00 代号：004

:::info 数据：
cls 从 0.5 上升到 0.7
命令行：
python train.py —weights weights/yolov5s.pt —cfg models/yolov5s6-bifpn_atten.yaml —data dataSet/clothes.yaml —hyp data/hyps/hyp.clothe009.yaml —batch-size 24 —epochs 70 —device 5 ::: �image.png
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: 0.5 # cls loss gain
cls: 0.7 # 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: 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3日下午 21:00 代号：005

:::info 数据：
cls 从 0.7 上升到 1.0
命令行：
python train.py —weights weights/yolov5s.pt —cfg models/yolov5s6-bifpn_atten.yaml —data dataSet/clothes.yaml —hyp data/hyps/hyp.clothe010.yaml —batch-size 36 —epochs 70 —device 6 ::: image.png
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: 0.7 # cls loss gain
cls: 1.0 # 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: 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3日下午 23:00 代号：006

:::info 数据：
cls 从 1.0 下降到 0.2
命令行：
python train.py —weights weights/yolov5s.pt —cfg models/yolov5s6-bifpn_atten.yaml —data dataSet/clothes.yaml —hyp data/hyps/hyp.clothe013.yaml —batch-size 24 —epochs 70 —device 1 ::: image.png
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: 1.0 # cls loss gain
cls: 0.2 # 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: 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3日下午 23:00 代号：007

:::info 数据：
cls 从 1.0 下降到 0.05
命令行：
python train.py —weights weights/yolov5s.pt —cfg models/yolov5s6-bifpn_atten.yaml —data dataSet/clothes.yaml —hyp data/hyps/hyp.clothe012.yaml —batch-size 36 —epochs 70 —device 5 ::: image.png
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: 1.0 # cls loss gain
cls: 0.05 # 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: 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3日下午 23:00 代号：008

:::info 数据：
cls 从 1.0 下降到 0.1
命令行：
python train.py —weights weights/yolov5s.pt —cfg models/yolov5s6-bifpn_atten.yaml —data dataSet/clothes.yaml —hyp data/hyps/hyp.clothe011.yaml —batch-size 36 —epochs 70 —device 4 ::: image.png
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: 1.0 # cls loss gain
cls: 0.1 # 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: 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

日志记录：10月3日下午12:30 代号：001

:::info 数据：
采用新的网络模型结果，增加了coord注意力模块
命令行：
python train.py —weights weights/yolov5s.pt —cfg models/yolov5s6-bifpn_atten.yaml —data dataSet/clothes.yaml —hyp data/hyps/hyp.clothe005.yaml —batch-size 36 —epochs 70 —device 3,4 ::: image.png
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: 0.01 # warmup initial bias lr
warmup_bias_lr: 0.055 # 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: 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: 4.0 # 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)