# DepthFM: Fast Monocular Depth Estimation with Flow Matching

Source: [DepthFM: Fast Monocular Depth Estimation with Flow Matching](<https://www.sota2.com/research/paper/68fa76541e209131cea6c714>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2025-10-26T03:51:28.908Z

Authors: Ming Gui, Johannes Schusterbauer, Ulrich Prestel, Pingchuan Ma, Dmytro Kotovenko, Olga Grebenkova, Stefan Andreas Baumann, Vincent Tao Hu, Bj\\"orn Ommer

Published: 2024-03-20T17:51:53.000Z

## Abstract

Current discriminative depth estimation methods often produce blurry artifacts, while generative approaches suffer from slow sampling due to curvatures in the noise-to-depth transport. Our method addresses these challenges by framing depth estimation as a direct transport between image and depth distributions. We are the first to explore flow matching in this field, and we demonstrate that its interpolation trajectories enhance both training and sampling efficiency while preserving high performance. While generative models typically require extensive training data, we mitigate this dependency by integrating external knowledge from a pre-trained image diffusion model, enabling effective transfer even across differing objectives. To further boost our model performance, we employ synthetic data and utilize image-depth pairs generated by a discriminative model on an in-the-wild image dataset. As a generative model, our model can reliably estimate depth confidence, which provides an additional advantage. Our approach achieves competitive zero-shot performance on standard benchmarks of complex natural scenes while improving sampling efficiency and only requiring minimal synthetic data for training.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Monocular Depth Estimation on ScanNet (Marigold)](<https://www.sota2.com/research/sota/monocular-depth-estimation-on-scannet-marigold.md>): Monocular Depth Estimation; ScanNet (Marigold); AbsRel: 6.6.

- [Relative Depth Estimation on DA-2K](<https://www.sota2.com/research/sota/relative-depth-estimation-on-da-2k.md>): Relative Depth Estimation; DA-2K; Accuracy: 85.8.

- [Monocular Depth Estimation on NYU-Depth V2 (test) (AbsRel, delta1)](<https://www.sota2.com/research/sota/monocular-depth-estimation-on-nyu-depth-v2-test-absrel-delta1.md>): Monocular Depth Estimation; NYU Depth V2 (test); AbsRel: 6.

- [Monocular Depth Estimation on 3D-Mirage](<https://www.sota2.com/research/sota/monocular-depth-estimation-on-3d-mirage.md>): Monocular Depth Estimation; 3D-Mirage; d\_cluster: 1020.

- [Monocular Depth Estimation on KITTI (AbsRel, delta1)](<https://www.sota2.com/research/sota/monocular-depth-estimation-on-kitti-absrel-delta1.md>): Monocular Depth Estimation; KITTI; AbsRel: 8.4.

- [Monocular Depth Estimation on DIODE](<https://www.sota2.com/research/sota/monocular-depth-estimation-on-diode.md>): Monocular Depth Estimation; DIODE; AbsRel: 22.4.

- [Monocular Depth Estimation on ETH3D](<https://www.sota2.com/research/sota/monocular-depth-estimation-on-eth3d.md>): Monocular Depth Estimation; ETH3D; AbsRel: 6.5.

- [Monocular Depth Estimation on NYU v2](<https://www.sota2.com/research/sota/monocular-depth-estimation-on-nyu-v2.md>): Monocular Depth Estimation; NYU V2; Delta 1 Acc: 95.6.

- [Monocular Depth Estimation on KITTI](<https://www.sota2.com/research/sota/monocular-depth-estimation-on-kitti.md>): Monocular Depth Estimation; KITTI; Abs Rel: 8.3.

- [Monocular Depth Estimation on KITTI (Eigen)](<https://www.sota2.com/research/sota/monocular-depth-estimation-on-kitti-eigen.md>): Monocular Depth Estimation; KITTI (Eigen); Abs Rel: 9.1.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2403.13788>)

[PDF](<https://arxiv.org/pdf/2403.13788.pdf>)
