yolov5-6.1版本的早停机制不起作用，设置patience=5，并不会触发早停。

[sc-guihub]

Search before asking

• I have searched https://github.com/ultralytics/yolov5/issues and did not find a similar report.

Project area

No response

Bug

yolov5-6.1版本的早停机制不起作用，设置patience=5，并不会触发早停。

Environment

yolov5-6.1版本的早停机制不起作用，设置patience=5，并不会触发早停。

Minimal Reproducible Example

No response

Additional

No response

Are you willing to submit a PR?

• Yes I'd like to help by submitting a PR!

[UltralyticsAssistant]

👋 Hello @sc-guihub, thank you for your report and for using YOLOv5 🚀! This is an automated response to help us triage your issue quickly—an Ultralytics engineer will also assist soon 🙂
Please visit our ⭐️ Tutorials to get started, where you can find quickstart guides for simple tasks like Custom Data Training all the way to advanced concepts like Hyperparameter Evolution.
If this is a 🐛 Bug Report, please provide a minimum reproducible example (MRE) to help us debug it.
If this is a custom training ❓ Question, please provide as much information as possible, including dataset image examples and training logs, and verify you are following our Tips for Best Training Results.

To help us investigate why early stopping with patience=5 isn’t triggering in YOLOv5 6.1, please add the following to this issue:
-Exact command used to train (full python train.py ... line) including where patience is set
-Your complete training log from start until the point you expect early stopping to trigger
-The YOLOv5 version/commit you’re using and how it was installed (clone + pip, etc.)
-Any custom changes to training (even small ones)
-A minimal dataset/config to reproduce (or a small subset + your data.yaml) as an MRE

Requirements

Python>=3.8.0 with all requirements.txt installed including PyTorch>=1.8. To get started:

git clone https://github.com/ultralytics/yolov5  # clone
cd yolov5
pip install -r requirements.txt  # install

Environments

YOLOv5 may be run in any of the following up-to-date verified environments (with all dependencies including CUDA/CUDNN, Python and PyTorch preinstalled):
-Notebooks with free GPU:
-Google Cloud Deep Learning VM. See GCP Quickstart Guide
-Amazon Deep Learning AMI. See AWS Quickstart Guide
-Docker Image. See Docker Quickstart Guide

Status

If this badge is green, all YOLOv5 GitHub Actions Continuous Integration (CI) tests are currently passing. CI tests verify correct operation of YOLOv5 training, validation, inference, export and benchmarks on macOS, Windows, and Ubuntu every 24 hours and on every commit.

[sc-guihub]

python -m torch.distributed.launch --nproc_per_node 4 train.py --weights yolov5m.pt --cfg models/yolov5m.yaml --data data/highway.yaml --epochs 300 --batch-size 64 --device 0,1,2,3 --patience 5

[pderrenger]

YOLOv5 的 --patience 早停是按“每个 epoch 验证集 fitness 是否有提升”来计数的，所以如果你的 data/highway.yaml 里没有配置 val:（或训练时关闭了验证），或者 fitness 仍在微小波动上升（哪怕很小），都不会触发早停；请先用最新 master（git pull）在单卡 --device 0 下复现一次并贴出包含每个 epoch 验证指标/fitness 的日志，以及 highway.yaml 里 train/val 路径配置，我们再一起确认是否是 v6.1 在 DDP 下的同步/停止信号问题。另可参考早停的判定逻辑说明见 What is Overfitting? (Early Stopping)。