Depth Map Prediction from a Single Image using a Multi-Scale Deep Network[NIPS 14]

One Line Summary

Depth estimation requiring integration of both global and local information from various cues, two deep network stacks are used to address this.

Motivation

The Depth estimation task is inherently ambiguous, with a large source of uncertainty coming from the overall scale, so a two scale coarse and fine predictions are used.

Detailed Summary

A new method that addresses this task by employing two deep network stacks: one that makes a coarse global prediction based on the entire image, and another that refines this prediction locally.
A scale-invariant error to help measure depth relations rather than scale. By leveraging the raw datasets as large sources of training data, this method achieves state-of-the-art results on both NYU Depth and KITTI.

Novelty and Contributions

Using a two scale deep network by first making predictions at coarse level and then using this to make predictions at fine level
Showing the state-of-the-art results on NYU Depth and KITTI.