SteeredMarigold: Steering Diffusion Towards Depth Completion of Largely Incomplete Depth Maps

TL;DR

SteeredMarigold is a training-free, zero-shot diffusion-based method that effectively completes largely incomplete depth maps, outperforming existing methods on standard datasets with high robustness.

Contribution

It introduces SteeredMarigold, a novel zero-shot depth completion approach that leverages diffusion models conditioned on sparse depth points, without requiring training.

Findings

Outperforms top methods on NYUv2 dataset with incomplete depth maps.

Achieves state-of-the-art results in scenarios with large missing areas.

Demonstrates robustness against depth map incompleteness.

Abstract

Even if the depth maps captured by RGB-D sensors deployed in real environments are often characterized by large areas missing valid depth measurements, the vast majority of depth completion methods still assumes depth values covering all areas of the scene. To address this limitation, we introduce SteeredMarigold, a training-free, zero-shot depth completion method capable of producing metric dense depth, even for largely incomplete depth maps. SteeredMarigold achieves this by using the available sparse depth points as conditions to steer a denoising diffusion probabilistic model. Our method outperforms relevant top-performing methods on the NYUv2 dataset, in tests where no depth was provided for a large area, achieving state-of-art performance and exhibiting remarkable robustness against depth map incompleteness. Our source code is publicly available at https://steeredmarigold.github.io.