Training from Scratch

This option is recommended for those training datasets on "novel" or niche projects, where there are no existing trained models with similar subjects to your data (e.g, astrophotography).

• Unsure if no trained model checkpoints exist for your target dataset, or use case? Search Roboflow Universe to look for similar datasets with trained models. Add the term "model" to the end of your query to filter for datasets with trained models.

To begin, navigate to the "Versions" page of your target dataset/project. Select the version you wish to train.

• You can only start a training job from a dataset version that has not already been trained with Roboflow Train (no green checkmark is present on the version).

Click "Start Training."

Next, choose the model architecture and then model size of the architecture then click "Continue."

• If you are training a Single-Label Classification, Multi-Label Classification, or Semantic Segmentation project, you will not have this Fast or Accurate option. For these project types, you can just click "Continue."

Proceed to select "Train from Scratch."

Now click "Start Training" to begin your training job. You will receive an email to the account email on file when the training job is complete.

Monitoring Your Training Job

You can monitor your training job's progress. The UI will show your machine starting up to begin training.

Viewing Your Training Results

Deploying Your Model

After training, your model is ready to be used for inference and embedded in a custom application! See the Inference Documentation page for all the options.