What is MMDNN?

MMDNN is a comprehensive, cross-framework solution to convert, visualize and diagnosis deep neural network models. With MMDNN, the gaps between different deep learning frameworks is narrowed or even eliminated. MMDNN is developed and supported by Microsoft and other community partners.

Framework Transferability

Enabling DNN conversion makes it possible to exploit DNN in other frameworks. MMDNN enables models to be trained in one framework and transferred to another for inference. MMDNN are currently supported in Caffe2, Apple Core ML, Keras, Microsoft Cognitive Toolkit, MXNet, and PyTorch, Tensorflow and there are connectors for many other common frameworks and libraries.

Model Visualibility

MMDNN makes it easier for researchers or engineers to observe the structure of neural networks. Visualization is rather important in designing the structure of Neural Network, and it will give you the insight of the model. Any converted model can be visualized by MMdnn Visualizer. MMdnn Visualizer now supports search the layer with their name!

Get Involved


Start experimenting today.
Check out our MMdnn Github, MMdnn Visualizer, and Wiki.


MMDNN is a open-source project. We encourage you to join the effort and contribute feedback, ideas, and code.

Check here for the Contribution Guideline.

Follow Us

Stay up to date with the latest MMDNN news.