Train

Our one-click training solution will give you a state of the art model hosted at an API endpoint customized for your dataset in no time.

Private Beta

Roboflow Train is currently in private beta. To get on the waiting list, click the Roboflow Train button on one of your datasets' versions. You will get a "Request Access" prompt. We will add more users to the beta soon!

Request Access via one of your dataset versions.

Overview

Roboflow is entering the AutoML game. Our goal is to be the easiest way to train and deploy a state of the art object detection model on your custom dataset. It's literally one click and we'll do the rest. When it's done training, you'll receive a hosted API where you can do inference on the web (or via your own application).

Pricing

Roboflow Train credits are included in our Roboflow Pro packages. Training a model costs one credit. It's that simple! Contact our sales team to upgrade to Roboflow Pro.

Usage

Choose preprocessing and augmentation settings as normal and then generate a version of your dataset. Click "Use Roboflow Train" then "Start Training" and we will train a model for you and return back an API you can use.

If you select a "Resize" preprocessing option we will train a model whose native input size is similar to your training data size.

To make your model faster, try exporting images in a smaller size; to make it more accurate, try a larger size. The maximum size we support is 2048x2048 (which is quite slow to train and inference against). If you don't select a Resize option, we default to 416x416.

Training

The larger your dataset (and the larger your outputted images), the longer it will take for your dataset to train. We will email you when it's finished.

Results

When your model has finished training, you can see the metrics on the dataset version page including mean average precision, precision, recall, and more.

Use Your Trained Model for Inference

After training, you will receive an API endpoint where you can receive predictions from your newly trained model.

See the Inference Documentation page

Next Steps

Try generating several versions of your dataset to see which preprocessing and augmentation options perform best and strike the best speed/accuracy tradeoff.