Research map / Image generation / Text-to-image diffusion
Camera Control for Text-to-Image Generation via Learning Viewpoint Tokens
Xinxuan Lu, Charless Fowlkes, Alexander C. Berg
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 |
|---|---|---|---|
| use factorized viewpoint encoding instead of Plucker rays (ablation) | Ours (Harmon) · Viewpoint test set (5,550 samples) | azimuth mean error (deg): 21.61 -> 18.11 | Improved |
| use factorized encoding instead of 12D camera matrix (ablation) | Ours (Harmon) · Viewpoint test set (5,550 samples) | azimuth mean error (deg): 24.44 -> 18.11 | Improved |
| use factorized encoding instead of sinusoidal encoding (ablation) | Ours (Harmon) · Viewpoint test set (5,550 samples) | azimuth mean error (deg): 60.90 -> 18.11 | Improved |
| add the rendered 3D subset to the training data (ablation) | Ours (Harmon) · Viewpoint test set (5,550 samples) | azimuth mean error (deg): 22.98 -> 18.11 | Improved |
| fine-tune the backbone instead of freezing it (ablation) | Ours (Harmon) · Viewpoint test set (5,550 samples) | azimuth mean error (deg): 40.19 -> 18.11 | Improved |
| add additional viewpoint tokens (ablation) | Ours (Harmon) · Viewpoint test set (5,550 samples) | azimuth mean error (deg): 18.11 -> 18.03 | No clear effect |
| swap Harmon backbone for SD3.5 (ablation) | Ours · Viewpoint test set (5,550 samples) | azimuth mean error (deg): 18.11 -> 12.85 | Mixed |
| swap Harmon backbone for SD2.1 (ablation) | Ours · Viewpoint test set (5,550 samples) | azimuth mean error (deg): 18.11 -> 19.16 | No clear effect |
| use global-context viewpoint tokens (ours) instead of Compass Control | Compass Control · Viewpoint test set (5,550 samples) | azimuth mean error (deg): 31.07 -> 18.11 | Improved |
| use ours instead of Compass Control on diverse objects | Compass Control · Diverse set (26 objects) | azimuth mean error (deg): 37.29 -> 19.06 | Improved |
| use ours instead of Compass Control on GenEval colors | Compass Control · GenEval | GenEval colors: 58.78 -> 76.60 | Improved |
| fine-tune Harmon with viewpoint tokens | Harmon · GenEval | GenEval single object: 99.90 -> 94.38 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Compass Control — Compass Control: Multi Object Orientation Control for Text-to-Image Generation (CVPR 2025) · better on 9 of 9 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- PreciseCam: Precise Camera Control for Text-to-Image Generation (CVPR 2025 · similarity 0.90)
- Learning Continuous 3D Words for Text-to-Image Generation (CVPR 2024 · similarity 0.89)
- Generative Photography: Scene-Consistent Camera Control for Realistic Text-to-Image Synthesis (CVPR 2025 · similarity 0.89)
- ViewDiff: 3D-Consistent Image Generation with Text-to-Image Models (CVPR 2024 · similarity 0.87)
- Learning Visual Generative Priors without Text (CVPR 2025 · similarity 0.87)
- LLM Blueprint: Enabling Text-to-Image Generation with Complex and Detailed Prompts (ICLR 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.
