Getting started

Quickstart

Our mission is to democratize the use of deep learning to solve computer vision challenges.
The main way we achieve this is by providing an automated neural network training service.
Here's how it works:

1. Create a project and hook up your dataset (or use one of ours)
  • Any S3-compatible object storage provider is supported. See the storage providers section for details.
  • The data must be annotated for supervised tasks. See the tasks section for annotation formats.
  • Our runners download your data for training; once it is done, the copy is deleted.
    We do not keep, much less share or sell your data!
2. Create a new model to train on that data
  • Select your desired hyperparameters. See the hyperparameters section to get a sense of them.
  • The model is queued, then assigned a runner to start training. See the training section for details.
  • While the model trains, you can check its live-updated metrics and compare them with other models within the project.
3. Use your trained models however you want
To interact with your models, you have 3 options:
  • Try the model directly in the web-app by drag-and-dropping a picture. The project is charged for the inference compute time.
  • Call the automatically-deployed REST endpoint. See the inference section for details. The project is charged for the inference compute time.
  • Export the model weights in PyTorch or ONNX format. The weights are not licensed, so they are yours to do whatever you want, including sharing them online or using them for commercial purposes. They're forever yours.