Supervisor

Suitable for

Abstract

Single image depth perception is a common task in computer vision which involves predicting the distance from the camera to the scene for each pixel in an image. This problem has broad applications in autonomous systems, virtual and augmented reality, as well as graphics. However, a significant challenge in single image depth prediction is the estimation of the absolute scene scale. Therefore, most depth estimation methods focus on predicting relative depth rather than absolute measurements [1]. This project aims to explore a new relative depth parameterisation based on the percentile ranking of depth values. The student will be expected to train a model using this parameterisation and evaluate its effectiveness by comparing its performance with existing methods on widely-used datasets such as NYUv2.

[1] https://arxiv.org/pdf/1907.01341v3.pdf

[2] https://cs.nyu.edu/~silberman/datasets/nyu_depth_v2.html

Pre-requisites: Suitable for those who have taken a course in machine learning