Dropout Not Working During Model Training

[Lingpheng]

Hi Ultralytics team,

I tried to add dropout when training my model, but it doesn’t seem to have any effect. Here is how I set it:

results = model.train(
    data=data_path,
    epochs=20,
    patience=15,
    batch=32,
    imgsz=640,
    device='0,1',
    seed=88,
    workers=cpu_cores,
    pretrained=False,
    optimizer='AdamW',
    lr0=0.001,
    lrf=0.1,
    dropout=0.3,
    project="runs",
    save_period=40,
    cls=0.8,
    close_mosaic=0,
)

To check whether the dropout parameter works, I also tried training with dropout=0:

results = model.train(
    data=data_path,
    epochs=20,
    patience=15,
    batch=32,
    imgsz=640,
    device='0,1',
    seed=88,
    workers=cpu_cores,
    pretrained=False,
    optimizer='AdamW',
    lr0=0.001,
    lrf=0.1,
    dropout=0,
    project="runs",
    save_period=40,
    cls=0.8,
    close_mosaic=0,
)

However, the resulting models from both runs are exactly the same.

Here's the comparison image showing identical performance:

Why isn't the dropout parameter having any effect?

Thanks in advance for your help!

[UltralyticsAssistant]

👋 Hello @Lingpheng, thank you for your interest in Ultralytics 🚀! We recommend a visit to the Docs for new users where you can find many Python and CLI usage examples and where many of the most common questions may already be answered.

If this is a 🐛 Bug Report, please provide a minimum reproducible example 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.

Join the Ultralytics community where it suits you best. For real-time chat, head to Discord 🎧. Prefer in-depth discussions? Check out Discourse. Or dive into threads on our Subreddit to share knowledge with the community.

Upgrade

Upgrade to the latest ultralytics package including all requirements in a Python>=3.8 environment with PyTorch>=1.8 to verify your issue is not already resolved in the latest version:

pip install -U ultralytics

Environments

YOLO 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 Ultralytics CI tests are currently passing. CI tests verify correct operation of all YOLO Modes and Tasks on macOS, Windows, and Ubuntu every 24 hours and on every commit.

This is an automated response. An Ultralytics engineer will also assist you here soon 😊.

[glenn-jocher]

Hi @Lingpheng! The dropout parameter in YOLO training settings is specifically designed for classification tasks, not detection tasks. Since you appear to be training a detection model (evidenced by parameters like cls=0.8 for classification loss weight), the dropout parameter has no effect on your model architecture. For detection models, YOLO uses other regularization techniques built into the architecture, which is why you're seeing identical results regardless of the dropout value.

[Lingpheng]

Hi @glenn-jocher, that makes perfect sense! So there's no need to manually add a dropout parameter for detection tasks, since YOLO already handles it internally. Thanks a lot for the explication.

[glenn-jocher]

You're absolutely right! YOLO detection models use other effective regularization techniques like weight decay, data augmentation (mosaic, mixup), and architectural designs that inherently prevent overfitting without needing explicit dropout layers. Glad that cleared things up for you! 😊