Check out frequently asked questions and
let us know if you still have questions.

  • Is it really free for public projects ?
    Yes, we want to do for deep learning, what Github did for open source software. If you share your data and models, you can use platform for free.
  • Once I train a model, what can I do with it?
    Once a model is trained by platform you can use it to predict unlabeled images. This can be done via csv download or using a prediction API.
  • Can I upload my own images?
    You can upload small collections (<10k images) right from your browser, larger collections can be imported from a GCS bucket.
  • How do you bill for usage?
    For private collections (public collections are always free), used for labeling or to do inference, we bill in 1k image increments. See our pricing page for details.
  • How fast are predictions?
    Our prediction API can be used in two modes, online or batch. The average latency for online is ˜ 100ms, the throughput on batch is typically 10k images per minute.
  • How many images do I need to hand label before I have a useful model?
    The answer depends on the characteristics of your dataset and the machine learning task, but typically 100–500 examples per class are enough to train a model.
  • Can I deploy on-premise?
    Enterprise customers can deploy models trained on platform for inference on-premise. Contact us for additional details.
  • Can I use platform to do something other than classification?
    Currently platform only supports multi-class classification. Other machine learning tasks like multi-label, object detection and segmentation will be available in the near future.
  • Who owns the data and the models?
    You always own the data. For public projects, platform owns the derivative products (including the labels, model weights, etc.).
    For private projects you also own any derivative products.