Research map / Image generation / Autoregressive and latent generation
Orchid: Image Latent Diffusion for Joint Appearance and Geometry Generation
Akshay Krishnan, Xinchen Yan, Vincent Casser, Abhijit Kundu
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 joint latent space (vs disentangled latents) (ablation) | Orchid · COCO text-conditioned | CLIP: 0.312 -> 0.316 | Improved |
| add joint latent space (vs disentangled latents) (ablation) | Orchid · COCO text-conditioned | LPIPS: 0.769 -> 0.764 | Hurt |
| add joint latent space (vs disentangled latents) (ablation) | Orchid · NYUv2 normals | accuracy within 11.25 deg: 54.8 -> 60.6 | Improved |
| add joint latent space (vs disentangled latents) (ablation) | Orchid · NYUv2 depth | delta1: 96.0 -> 96.9 | Improved |
| add VAE distillation loss (ablation) | Orchid · COCO text-conditioned | CLIP: 0.309 -> 0.316 | Improved |
| add VAE distillation loss (ablation) | Orchid · NYUv2 normals | accuracy within 11.25 deg: 54.6 -> 60.6 | Improved |
| add VAE distillation loss (ablation) | Orchid · NYUv2 depth | delta1: 96.1 -> 96.9 | Improved |
| add pseudo depth-normal labels on text-image data (ablation) | Orchid · NYUv2 normals | accuracy within 11.25 deg: 57.9 -> 60.6 | Improved |
| add pseudo depth-normal labels on text-image data (ablation) | Orchid · NYUv2 depth | delta1: 96.6 -> 96.9 | Improved |
| joint diffusion instead of Diff+Diff or Diff+FF pipelines | Orchid · inference time | seconds per image: 4.2 -> 1.2 | Improved |
| joint diffusion instead of color diffusion plus feed-forward depth/normal models | Orchid · inference time | seconds per image: 1.3 -> 1.2 | Improved |
| joint color-depth-normal latent vs separate depth and normal models | Orchid · KITTI depth-normal consistency | inconsistency error: 0.144 -> 0.082 | Improved |
Closest papers by meaning
- LT3SD: Latent Trees for 3D Scene Diffusion (CVPR 2025 · similarity 0.87)
- Color Alignment in Diffusion (CVPR 2025 · similarity 0.87)
- HumanNorm: Learning Normal Diffusion Model for High-quality and Realistic 3D Human Generation (CVPR 2024 · similarity 0.87)
- RichDreamer: A Generalizable Normal-Depth Diffusion Model for Detail Richness in Text-to-3D (CVPR 2024 · similarity 0.86)
- In-N-Out: Lifting 2D Diffusion Prior for 3D Object Removal via Tuning-Free Latents Alignment (NeurIPS 2024 · similarity 0.86)
- Geometry Image Diffusion: Fast and Data-Efficient Text-to-3D with Image-Based Surface Representation (ICLR 2025 · similarity 0.86)
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.
