Frequently Asked Questions
How do I get started using DeepCell?
If you’d like to use our pretrained models to segment your own data, you can use the predict page. The predict page allows you to easily upload your images with a drag and drop interface, select the most appropriate model, and get predictions back all without needing to install any software.
If you’d like to train your own models, check out deepcell-tf. If you’d like to annotate your data, you can use the DeepCell Label tool, available via our website or from the GitHub repository.
What does this error message mean?
Invalid image shape
The image provided is not compatible with the model. Check that the channels of the input image match the expected model output, and that the dimensions of the image match the model (i.e. 2D images or 3D movies).
Input only has X channels but channel Y was declared as an input channel.
An RGB channel was specified but that channel does not exist in the input image.
Input image is larger than the maximum supported image size of (M, N).
Your input image is too big! Try cropping the image and uploading the crops separately.
Can I add my own models?
Yes! deepcell.org is an instance of the kiosk-console which is fully extensible and serves models from a cloud bucket using TensorFlow Serving.
For more information on creating and customizing your own instance of the kiosk-console, please check out its docs.
Can you help me annotate my data?
Yes! Our training data was created using DeepCell Label, a tool for creating segmentation masks for images. DeepCell Label is an open-source web application that can integrate with crowd-sourcing platforms.
If you have any questions or interest in collaborating on the data annotation process, please make a new Issue on the repository issue page.
Where can I get help?
For an overview of the DeepCell ecosystem, please see the About page and our introductory docs.
If you would like to report a bug or ask a question, please open a new issue on the issues page.