Overview / Usage

This project aims to experiment with masking images based on estimated depth values from the Intel MiDaS model. This is being done to see if there are performance gains on low powered computer vision applications where there is a set of relevant pixels that need to be analyzed for classification tasks.

The problem that I'm trying to address with this project is that current low powered computer vision applications and implementations are less performant than their counterpart. As the demand for computer vision models on mobile and embedded systems increases, current and future models need to be able to perform regardless of power constraints.

Repository

https://github.com/NicholasSynovic/image-depth-masking