11月 5日 炼丹

李志颖

yolov5s

:::tips

mAP0.5 22.5

mAP 11.4

::: :::info 数据：采用yolov5s来训练模型
命令行：
python train.py —weights weights/yolov5s.pt —cfg models/yolov5s.yaml —hyp data/hyps/hyp.scratch.yaml —data dataSet/clothes.yaml —batch-size 16 —epochs 400 —device 4,5 ::: :::info 命令行：python val.py —data dataSet/clothes.yaml —weights runs/train/exp204/weights/best.pt —iou-thres 0.45 —device 0 —verbose —save-txt :::

yolov5m

:::tips

mAP0.5 24.3

mAP 12.9

::: :::info 数据：采用yolov5m来训练模型
命令行：
python train.py —weights weights/yolov5m.pt —cfg models/yolov5m.yaml —hyp data/hyps/hyp.scratch.yaml —data dataSet/clothes.yaml —batch-size 16 —epochs 400 —device 4,5 ::: :::info 命令行：python val.py —data dataSet/clothes.yaml —weights runs/train/exp206/weights/best.pt —iou-thres 0.45 —device 0 —verbose —save-txt :::

yolov5l

:::tips

mAP0.5 28.8

mAP 15.3

::: :::info 数据：采用yolov5l来训练模型
命令行：
python train.py —weights weights/yolov5l.pt —cfg models/yolov5l.yaml —hyp data/hyps/hyp.best1.yaml —data dataSet/clothes.yaml —batch-size 16 —epochs 400 —device 3,4 ::: :::info 命令行：python val.py —data dataSet/clothes.yaml —weights runs/train/exp212/weights/best.pt —iou-thres 0.45 —device 0 —verbose —save-txt :::

yolov5x

:::tips

mAP0.5 30.6

mAP 17.1

::: :::info 数据：采用yolov5x来训练模型
命令行：
python train.py —weights weights/yolov5x.pt —cfg models/yolov5x.yaml —hyp data/hyps/hyp.scratch01.yaml —data dataSet/clothes.yaml —batch-size 18 —epochs 400 —device 0,2,5 :::

:::info 命令行：python val.py —data dataSet/clothes.yaml —weights runs/train/exp240/weights/best.pt —iou-thres 0.45 —device 5 —verbose —save-txt :::

yolov5s6+Four Head+Atten

:::tips

mAP0.5 29.8

mAP 14.2

::: :::info 数据：采用yolov5s6_atten.yaml来训练模型
命令行：
python train.py —weights weights/yolov5s6.pt —cfg models/yolov5s6_atten.yaml —hyp data/hyps/hyp.best.yaml —data dataSet/clothes.yaml —batch-size 16 —epochs 400 —device 1,2 ::: :::info 命令行：python val.py —data dataSet/clothes.yaml —weights runs/train/exp202/weights/best.pt —iou-thres 0.45 —device 0 —verbose —save-txt :::

yolov5s6+Four Head+Bifpn

:::tips

mAP0.5 28.4

mAP 15.6

::: :::info 数据：采用yolov5s6-bifpn.yaml来训练模型
命令行：
python train.py —weights weights/yolov5s6.pt —cfg models/yolov5s6-bifpn.yaml —hyp data/hyps/hyp.best3.yaml —data dataSet/clothes.yaml —batch-size 16 —epochs 400 —device 1,5 ::: :::info 命令行：python val.py —data dataSet/clothes.yaml —weights runs/train/exp219/weights/best.pt —iou-thres 0.45 —device 0 —verbose —save-txt :::

Our Model

:::tips

mAP0.5 31.5

mAP 15.9

::: :::info 数据：Mixup由0转变为了0.243
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 1,5 ::: :::info 命令行：python val.py —data dataSet/clothes.yaml —weights runs/train/exp233/weights/best.pt —iou-thres 0.45 —device 0 —verbose —save-txt :::

Centernet (Resnet-50)

:::tips mAP: 10.79
Flops: 23.2GMac
Params: 32.67M :::

yolov3

:::tips mAP: 6.81
Flops: 19.62 GMac
Params: 62.12 M :::