Research map / Image generation / Text-to-image diffusion
Marigold: Affordable Adaptation of Diffusion-Based Image Generators for Image Analysis
Bingxin Ke, Kevin Qu, Tianfu Wang, Nando Metzger, Shengyu Huang, Bo Li, Anton Obukhov, Konrad Schindler B
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 |
|---|---|---|---|
| add Virtual KITTI to HyperSim training data (ablation) | Marigold-Depth · KITTI (depth ablation) | AbsRel: 13.7 -> 11.3 | Improved |
| add Virtual KITTI to HyperSim training data (ablation) | Marigold-Depth · NYUv2 (depth ablation) | AbsRel: 5.7 -> 5.6 | Improved |
| add HyperSim to Virtual KITTI training data (ablation) | Marigold-Depth · NYUv2 (depth ablation) | AbsRel: 13.9 -> 5.6 | Improved |
| use trailing DDIM timesteps instead of leading (ablation) | Marigold-Depth · depth benchmarks | NFE needed: 50 steps -> 1 step | Improved |
| increase test-time ensemble from 1 to 10 predictions (ablation) | Marigold-Depth · NYUv2 | AbsRel: ~8% | Improved |
| increase test-time ensemble from 1 to 20 predictions (ablation) | Marigold-Depth · NYUv2 | AbsRel: ~9.5% | Improved |
| use model v1.1 with 10x4 ensemble and steps instead of 1x1 (ablation) | Marigold-Depth v1.1 · NYUv2 | AbsRel: 5.9 -> 5.5 | Improved |
| run in fp16 | Marigold v1.1 (1x1) · inference time | time (s): 0.568 -> 0.274 | Improved |
| use TAESD VAE with fp16 | Marigold v1.1 (1x1) · inference time | time (s): 0.274 -> 0.082 | Improved |
| use TAESD VAE with fp16 | Marigold v1.1 (1x1) · KITTI | AbsRel: 11.0 -> 12.4 | Hurt |
| increase denoising steps from 1 to 4 (ablation) | Marigold-Normals · normals benchmarks | accuracy | Improved |
| increase ensemble beyond 10 predictions (ablation) | Marigold-Normals · ScanNet and DIODE | accuracy | No clear effect |
Closest papers by meaning
- Repurposing Diffusion-Based Image Generators for Monocular Depth Estimation (CVPR 2024 · similarity 0.90)
- Lotus: Diffusion-based Visual Foundation Model for High-quality Dense Prediction (ICLR 2025 · similarity 0.88)
- Boosting Generative Image Modeling via Joint Image-Feature Synthesis (NeurIPS 2025 · similarity 0.88)
- ZoomLDM: Latent Diffusion Model for Multi-scale Image Generation (CVPR 2025 · similarity 0.88)
- SaRA: High-Efficient Diffusion Model Fine-tuning with Progressive Sparse Low-Rank Adaptation (ICLR 2025 · similarity 0.87)
- A Simple Approach to Unifying Diffusion-based Conditional Generation (ICLR 2025 · 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.
