Research map / 3D vision / 3D generation
Dream-to-Recon: Monocular 3D Reconstruction with Diffusion-Depth Distillation from Single Images
Philipp Wulff, Felix Wimbauer, Dominik Muhle, Daniel Cremers
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| train the reconstruction model on synthetic single-image geometry instead of multi-view supervision | Dream-to-Recon distilled model · KITTI-360 | IEacc: 0.71 vs BTS 0.69 | Improved |
| add depth supervision to BTS baseline | BTS · KITTI-360 | IErec: 0.64 -> 0.66 | No clear effect |
| use full conditioning (RGB, depth, mask, CLIP) in the view completion model (ablation) | VCM · KITTI-360 | PSNR: 20.197 -> 23.908 | Improved |
| finetune VCM on Waymo (ablation) | VCM · Waymo | PSNR: 26.963 -> 28.781 | Improved |
| use camera rig with 8 poses instead of random orbit with 4 (ablation) | scene synthesis · KITTI-360 | IErec: 0.596 -> 0.729 | Improved |
| add depth rendering loss Ldepth (ablation) | distillation · KITTI-360 | IErec: 0.443 -> 0.706 | Improved |
| add occupancy loss Locc (ablation) | distillation · KITTI-360 | IEacc: IEacc 0.699 -> 0.705; IErec 0.757 -> 0.706 | Mixed |
| add predicted uncertainty to balance losses (ablation) | distillation · KITTI-360 | IEacc: 0.697 -> 0.705 | Improved |
| use depth-gradient occlusion detection instead of optical flow (ablation) | scene synthesis · KITTI-360 | occlusion detection quality (qualitative) | Improved |
| distill synthetic scenes into feed-forward model | Dream-to-Recon · KITTI-360 | Oacc: 0.93 -> 0.90 | Hurt |
Closest papers by meaning
- Diffusion Models for Monocular Depth Estimation: Overcoming Challenging Conditions (ECCV 2024 · similarity 0.89)
- Repurposing Diffusion-Based Image Generators for Monocular Depth Estimation (CVPR 2024 · similarity 0.89)
- Coherent 3D Scene Diffusion From a Single RGB Image (NeurIPS 2024 · similarity 0.88)
- DepthFM: Fast Generative Monocular Depth Estimation with Flow Matching (AAAI 2025 · similarity 0.88)
- Decompositional Neural Scene Reconstruction with Generative Diffusion Prior (CVPR 2025 · similarity 0.87)
- BetterDepth: Plug-and-Play Diffusion Refiner for Zero-Shot Monocular Depth Estimation (NeurIPS 2024 · similarity 0.87)
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.
