基于YOLOv8的手部检测（2）- 模型训练、结果分析和超参数优化

paradoxjun

已于 2024-08-17 14:02:39 修改

阅读量1k

点赞数 29

分类专栏：目标检测手势识别文章标签： YOLO 目标检测计算机视觉人工智能深度学习

于 2024-08-13 16:14:36 首次发布

本文链接：https://blog.csdn.net/qq_40387714/article/details/141164131

版权

目标检测同时被 2 个专栏收录

13 篇文章 5 订阅

订阅专栏

手势识别

7 篇文章 1 订阅

订阅专栏

2.1 HaGRID上训练结果的准确率分析

2.1 HaGRID上训练结果的LOSS分析

前言

对YOLOv8手部检测模型进行训练，并分析训练结果，从而调优训练超参数。

手部检测数据集：基于YOLOv8的手部检测（1）- 手部数据集获取（数据集下载、数据清洗、处理与增强）

1.训练参数设置

1.1 data.yaml

这里没有控制每一轮的正负样本比和每一轮的长度，只加入额外负样本数，将一个epoch的训练数据量控制为36W张图片。因为原始训练集正样本有22W张，正负样本比接近3：2，已经算比较合理的范围了。

# path: /path/to/datasets
train: [""]
val: [""]
 
# Add negative image ----------------------------------------------------------
negative_setting:
  neg_ratio: 0    # 小于等于0时，按原始官方配置训练，大于0时，控制正负样本。
  use_extra_neg: True
  extra_neg_sources: {"./datasets/hagrid/yolo_pose_neg/train/images": 42413,
                      # "./datasets/negative_dataset/neg_hand/train": 50000
                      }
  fix_dataset_length: 0    # 是否自定义每轮参与训练的图片数量
 
# number of classes -----------------------------------------------------------
nc: 1
 
# Classes ---------------------------------------------------------------------
names:
  0: hand

1.2 setting.yaml

override模型参数注意：需要把关闭马赛克增强置为0（close_mosaic: 0）；将马赛克增强概率设置为0.9（mosaic: 0.9）；开启多尺度会提高训练效果，但很消耗内存（multi_scale: True）。

# Train settings -------------------------------------------------------------------------------------------------------
task: detect            # (str) YOLO task, i.e. detect, segment, classify, pose
mode: train             # (str) YOLO mode, i.e. train, val, predict, export, track, benchmark
data: ./data.yaml       # (str, optional) path to data file, i.e. coco8.yaml
epochs: 500             # (int) number of epochs to train for
batch: 80               # (int) number of images per batch (-1 for AutoBatch)
imgsz: 640              # (int | list) input images size as int for train and val modes, or list[w,h] for predict and export modes
patience: 300           # (int) epochs to wait for no observable improvement for early stopping of training
device: 4               # (int | str | list, optional) device to run on, i.e. cuda device=0 or device=0,1,2,3 or device=cpu
project: ./             # (str, optional) project name
name:                   # (str, optional) experiment name, results saved to 'project/name' directory
single_cls: True        # (bool) train multi-class data as single-class
close_mosaic: 0         # (int) disable mosaic augmentation for final epochs (0 to disable)
multi_scale: True       # (bool) Whether to use multiscale during training

# Hyperparameters ------------------------------------------------------------------------------------------------------
lr0: 0.01               # (float) initial learning rate (i.e. SGD=1E-2, Adam=1E-3)
lrf: 0.01               # (float) final learning rate (lr0 * lrf)
box: 8.0                # (float) box loss gain, default: 7.5
cls: 0.5                # (float) cls loss gain (scale with pixels), default: 0.5
dfl: 2.0                # (float) dfl loss gain, default: 1.5
degrees: 0.0            # (float) image rotation (+/- deg)
translate: 0.1          # (float) image translation (+/- fraction)
scale: 0.5              # (float) image scale (+/- gain)
fliplr: 0.5             # (float) image flip left-right (probability)
mosaic: 0.9             # (float) image mosaic (probability)