Research

Paper

TESTING March 11, 2026

Need for Speed: Zero-Shot Depth Completion with Single-Step Diffusion

Authors

Jakub Gregorek, Paraskevas Pegios, Nando Metzger, Konrad Schindler, Theodora Kontogianni, Lazaros Nalpantidis

Abstract

We introduce Marigold-SSD, a single-step, late-fusion depth completion framework that leverages strong diffusion priors while eliminating the costly test-time optimization typically associated with diffusion-based methods. By shifting computational burden from inference to finetuning, our approach enables efficient and robust 3D perception under real-world latency constraints. Marigold-SSD achieves significantly faster inference with a training cost of only 4.5 GPU days. We evaluate our method across four indoor and two outdoor benchmarks, demonstrating strong cross-domain generalization and zero-shot performance compared to existing depth completion approaches. Our approach significantly narrows the efficiency gap between diffusion-based and discriminative models. Finally, we challenge common evaluation protocols by analyzing performance under varying input sparsity levels. Page: https://dtu-pas.github.io/marigold-ssd/

Metadata

arXiv ID: 2603.10584
Provider: ARXIV
Primary Category: cs.CV
Published: 2026-03-11
Fetched: 2026-03-12 04:21

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2603.10584v1</id>\n    <title>Need for Speed: Zero-Shot Depth Completion with Single-Step Diffusion</title>\n    <updated>2026-03-11T09:40:03Z</updated>\n    <link href='https://arxiv.org/abs/2603.10584v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2603.10584v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>We introduce Marigold-SSD, a single-step, late-fusion depth completion framework that leverages strong diffusion priors while eliminating the costly test-time optimization typically associated with diffusion-based methods. By shifting computational burden from inference to finetuning, our approach enables efficient and robust 3D perception under real-world latency constraints. Marigold-SSD achieves significantly faster inference with a training cost of only 4.5 GPU days. We evaluate our method across four indoor and two outdoor benchmarks, demonstrating strong cross-domain generalization and zero-shot performance compared to existing depth completion approaches. Our approach significantly narrows the efficiency gap between diffusion-based and discriminative models. Finally, we challenge common evaluation protocols by analyzing performance under varying input sparsity levels. Page: https://dtu-pas.github.io/marigold-ssd/</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.CV'/>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.RO'/>\n    <published>2026-03-11T09:40:03Z</published>\n    <arxiv:primary_category term='cs.CV'/>\n    <author>\n      <name>Jakub Gregorek</name>\n    </author>\n    <author>\n      <name>Paraskevas Pegios</name>\n    </author>\n    <author>\n      <name>Nando Metzger</name>\n    </author>\n    <author>\n      <name>Konrad Schindler</name>\n    </author>\n    <author>\n      <name>Theodora Kontogianni</name>\n    </author>\n    <author>\n      <name>Lazaros Nalpantidis</name>\n    </author>\n  </entry>"
}