Train state-of-the-art computer vision models on your datasets - no ML expertise or GPU setup required. Roboflow manages the training infrastructure and hosts the resulting models for deployment.
Choose & Train
Supported Models
Every model you can train, upload, or deploy with Roboflow, and where each runs.
Train a Model
Start a training job from the dashboard, API, SDK, or CLI.
Roboflow Instant
A model trained automatically from your first approved annotations.
Neural Architecture Search
Discover and fine-tune novel architectures.
Roboflow Rapid
Build a working model from a small sample of data in minutes.
Train from a Checkpoint
Start training from an existing Roboflow Universe model.
You can also train from a third-party checkpoint on Azure Vision or Google Cloud.
Evaluate & Understand
Evaluate Trained Models
Evaluation runs, mAP, and confusion matrices.
Training Results
Review metrics and graphs from a completed training job.
View Trained Models
Browse the models trained in your workspace.
Training Resolutions
See the input resolution used by each model type.
Versions, Trainings, and Models
How datasets, training jobs, and models relate.
Model IDs
How models are identified when you deploy or run inference.
Trained models deploy through the Serverless Cloud API, Dedicated Deployments, or self-hosted Inference.