zero-to-hero

Create and deploy to production a simple neural network for Computer Vision

Tools Used

JAX Library for computing gradients, performing tensor operations and scheming the segmentation model
Wandb for metrics and training tools
MLflow for deploying and compiling the model for production
Gradio for interactive user-experience platform within an online platform (Data-ICMC Website).

Datasets to consider

LabelMe 12 50k
City-Scapes
CamVid

References

DSNet-Fast
DSNet

First Model

The first model available and deployed considers a simple 2D image, taken from the MNIST Dataset. Reefer to High-Performance Neural Networks for Visual Object Classification for further details on the dataset.

The CI/CD process will use the default Github pipeline using the available Github Actions features. The training process will use the MLFLow framework, to cather and track the necessary metrics and log accordingly. Reefer to the docs for further details.