GUI Prototype to Visualize the loss function
- 0 Collaborators
GUI for Deep learning training, inference with live loss monitoring. ...learn more
Project status: Under Development
Intel Technologies
oneAPI,
Intel FPGA,
Movidius NCS,
OpenVINO,
Intel Python
Overview / Usage
I am a Deep learning applications developer and am familiar with the API's. But if a company has to opt for a product based on deep learning they have to hire new talent and need to spend a lot of resources. So, if we are able to build a GUI that can help them train their models then it would help them save a lot of resources and the product development time would reduce drastically.
This idea is not new. Already there is DIGITS by NVIDIA but this only works on a GPU perfectly and is problematic with a CPU. Also it cannot provide support for other devices like VPU , FPGAs, etc.
Our idea is to provide a Deep Learning engine powered by the Intel Open Vino kit and can help us in building a complete computer vision pipeline via a simple and easy to use GUI.
Methodology / Approach
Our methods were quite simple. In this prototype we build a CNN that classifies the MNIST dataset. Then we use OOPs concepts to create modular files that we used to visualize the loss function dynamically. At the end we converted the torch model to ONNX and then fed it to the model optimizer and got the IR files. Which we used to do a sample inference.
Currently we tested this prototype on VPU, CPU & GPU. The GUI and other entities still need improvement which we are working on.
Technologies Used
Intel Dev Cloud , Python - Jupyter Notebook , NCS2