Research map / Image generation / Fast diffusion sampling
The Surprising Effectiveness of Diffusion Models for Optical Flow and Monocular Depth Estimation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use DDVM instead of the strongest listed zero-shot RAFT baseline | Optical flow estimation · Sintel.final (zero-shot) | AEPE: 2.28 -> 2.00 | Helped |
| Use DDVM instead of FlowFormer | Optical flow estimation · Sintel.final (fine-tuned) | AEPE: 2.18 -> 2.48 | Hurt |
| Use DDVM instead of RAFT-it | Optical flow estimation · KITTI (fine-tuned public benchmark) | Fl-all: 4.31% -> 3.26% | Helped |
| Use DDVM instead of AiT-P | Monocular depth estimation · NYU-Depth-v2 | REL: 0.076 -> 0.074 | Helped |
| Add both in-filling and step-unrolling (ablation) | DDVM flow fine-tuning · KITTI val | AEPE: - -> 1.47 | Helped |
| Add coarse-to-fine refinement (ablation) | DDVM optical flow · Sintel.final zero-shot | AEPE: 2.12 -> 2.00 | Helped |
| Add FlyingThings, Kubric, and TartanAir to AutoFlow pretraining (ablation) | DDVM optical flow · Sintel.final zero-shot | AEPE: 2.55 -> 2.00 | Helped |
| Add ScanNet to SceneNet RGB-D pretraining (ablation) | DDVM monocular depth · NYU-Depth-v2 | REL: 0.089 -> 0.075 | Helped |
| Use L1 rather than L2 denoiser loss (ablation) | DDVM monocular depth · NYU-Depth-v2 | REL: 0.085 -> 0.075 | Helped |
| Add self-supervised pre-training (ablation) | DDVM monocular depth · NYU-Depth-v2 | REL: 0.081 -> 0.075 | Helped |
| Replace step-unrolling with self-conditioning (ablation) | DDVM monocular depth · NYU-Depth-v2 | REL: 0.076 -> 0.082 | Hurt |
| Reduce inference to 24 denoising steps (ablation) | DDVM monocular depth · NYU-Depth-v2 | REL: 0.075 -> 0.104 | Hurt |
Closest papers by meaning
- Repurposing Diffusion-Based Image Generators for Monocular Depth Estimation (CVPR 2024 · similarity 0.91)
- DepthFM: Fast Generative Monocular Depth Estimation with Flow Matching (AAAI 2025 · similarity 0.90)
- Hierarchical Flow Diffusion for Efficient Frame Interpolation (CVPR 2025 · similarity 0.89)
- FlowDiffuser: Advancing Optical Flow Estimation with Diffusion Models (CVPR 2024 · similarity 0.89)
- DiffSF: Diffusion Models for Scene Flow Estimation (NeurIPS 2024 · similarity 0.89)
- Jasmine: Harnessing Diffusion Prior for Self-supervised Depth Estimation (NeurIPS 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.