Train a Model

You can start a training job either in the Roboflow platform or using the Python SDK, using the instructions below.

Python SDK

To schedule a training job via the Python SDK, use the train() method on the version of your dataset for which you would like to train your model. Note: this code initiates training on the Roboflow platform asynchronously, and the code will finish executing before training completes.

import roboflow

rf = roboflow.Roboflow(api_key=YOUR_API_KEY_HERE)

# List all projects for your workspace
workspace = rf.workspace()

# get a project
project = rf.workspace().project("PROJECT_ID")

# Create a new version with custom preprocessing and augmentation
new_version = project.generate_version(
    preprocessing={
        "auto-orient": True,
        "resize": {"width": 640, "height": 640, "format": "Stretch to"},
        "grayscale": False
    },
    augmentation={
    }
)
version = project.version(new_version)

# Train on the version with specific training parameters
model = version.train(
    speed="fast",          # Options: "fast" (default) or "accurate" (paid feature)
    checkpoint=None,       # Use a specific checkpoint to continue training
    plot_in_notebook=False # Visualize training progress (for notebooks)
)