(Formally known as TensorFlow GUI.)
Neural studio is a rapid prototyping tool for deep learning applications. It provides and easy to use and intuitive UI for users ranging from beginners to professionals.
Student : Viraj Patel
Mentor : Monjoy Saha
Organization : Department of Biomedical Informatics (BMI), Emory University School of Medicine
☑️ : Implementation Partial
✅ : Implementation Complete
✅ Re-design and re-implement UI in React and Python based backend framework and move the application to the web.
✅ Removing dependencies like konvajs and creating a Graph editor from scratch.
✅ Add support for keras APIs.
✅ Ready to use pre-trained models like VGG16, ResNet, etc. These can be used for transfer learning.
✅ Create an in-browser training and testing environment.
✅ Develop utilities to monitor training process, monitor live loss and visualize them using line plots.
✅ Add support for custom functions as nodes that can be used to inject code in graph.
☑️ Develop an Inference engine for trained models to run models and utility to visualize outputs for simpler models like classification, regression segmentation etc.
☑️ Add custom data loaders and a dataset viewer for different input type of data. Eg. images, text files.
☑️ Add support for saving , loading and sharing training models in different formats such as .pb , weights and json files.
✅ Add preprocessed open source datasets from different domains.
These are features that i proposed as part of my GSoC proposal. I have also added severl more features like cross platform distribution using PyPi, Public Inference APIs for trained models and Workspace management.
Links
Tutorials
Contribution Guide.
coming soon...