Default Augmentation when using the Train Mode

[NiklasHansen1993]

Hello everyone,
i'm very new to using Objectdetection and anything to do with coding in general. I've been training a few models on datasets consisting of Blood smear images in order to detect white blood cells. When starting the training with IMGSZ = 640
BATCH = 16
EPOCHS = 300
PATIENCE = 50
i get a long list of values for different augmentation arguments - does YOLO automatically apply augmentation for training if nothing else is specified? And is this reccommended?

Thank you in advance, Niklas

[UltralyticsAssistant]

👋 Hello @NiklasHansen1993, thank you for your interest in Ultralytics 🚀! This is an automated response to help you get started, and an Ultralytics engineer will also assist you soon 😊. 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 (like your augmentation/training-settings 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. In particular, please paste the portion of your training output where the augmentation arguments are printed, plus your exact train command/script and ultralytics version.

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.

[glenn-jocher]

Yes—if you don’t specify augmentation args, Ultralytics YOLO will train with the default augmentation settings (and prints them at startup); that’s generally recommended as a baseline, but for microscopy/blood-smear data you may want to dial back “scene-changing” transforms (especially mosaic and strong geometry) if they create unrealistic samples—see the YOLO augmentation guide and the training augmentation arguments table.

from ultralytics import YOLO

model = YOLO("yolo26n.pt")
model.train(
    data="your_dataset.yaml",
    imgsz=640,
    batch=16,
    epochs=300,
    patience=50,
    mosaic=0.0,      # often worth trying for microscopy
    mixup=0.0,
    degrees=0.0,
    translate=0.0,
    shear=0.0,
    perspective=0.0,
    # keep mild color aug if staining/illumination varies:
    hsv_h=0.01,
    hsv_s=0.2,
    hsv_v=0.2,
)

[NiklasHansen1993]

perfect, thank you so much, i will try that! I'm also trying to test the models performance when training with a smaller and smaller amount of training data - do you have any tips for that regarding the hyperparameters? Or are there legit ways of using Augmentations to enlarge a small dataset that only consists of 10 - 50 pictures by copying part of the training data and using augmentations on these copies?

[glenn-jocher]

You don’t need to physically copy images to “grow” the dataset—Ultralytics YOLO applies augmentations on-the-fly each epoch, so a 10–50 image set effectively produces many unique variants during training; for low-data experiments, keep one fixed held-out test/val split (don’t shrink it each run) and expect high variance, so train a few repeats and prefer a smaller model like yolo26n.pt to reduce overfitting, then only add augmentations that still look like real blood-smear images (more isn’t always better). If you want stronger/medical-specific variation beyond the built-ins, you can pass custom Albumentations transforms via the Python API as shown in the Albumentations integration guide (this augments in-memory during training; it’s not creating saved copies), and the general idea is summarized in the data augmentation glossary.