Research map / Image generation / Text-to-image diffusion
OPRO: Orthogonal Panel-Relative Operators for Panel-Aware In-Context Image Generation
Sanghyeon Lee, Minwoo Lee, Euijin Shin, Kangyeol Kim, Seunghwan Choi, Jaegul Choo
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 OPRO to LoRA (ComRoPE backbone) | LoRA r=8 · Compositional reasoning 4x4 | accuracy (%): 29.20 -> 47.20 | Improved |
| add OPRO to LoRA (RoPE backbone) | LoRA r=8 · Compositional reasoning 3x3 | accuracy (%): 36.20 -> 42.00 | Improved |
| add OPRO to LoRA (LieRE backbone) | LoRA r=8 · Compositional reasoning 2x2 | accuracy (%): 58.10 -> 58.10 | No clear effect |
| add OPRO to LoRA (APE absolute encoding) | LoRA r=8 · Compositional reasoning 4x4 | accuracy (%): 19.50 -> 24.20 | Improved |
| add isometry and same-panel invariance via OPRO instead of additive panel bias (APB) (ablation) | LoRA ViT-B · Compositional reasoning 3x3 | accuracy (%): 35.70 -> 42.00 | Improved |
| tie query and key operators (same-panel invariance) instead of Asym-OPRO (ablation) | OPRO · Compositional reasoning 3x3 | accuracy (%): 39.70 -> 42.00 | Improved |
| add zero initialization of OPRO (ablation) | OPRO · Compositional reasoning 3x3 | accuracy (%): 38.60 -> 42.00 | Improved |
| add OPRO on top of LoRA vs LoRA baseline (additive bias reduces accuracy) (ablation) | LoRA · Compositional reasoning 3x3 | accuracy (%): 36.20 -> 35.70 | Hurt |
| add OPRO to ICEdit | ICEdit · MagicBrush | L1: 0.1189 -> 0.0781 | Improved |
| add OPRO to UNO | UNO · MagicBrush | DINO: 0.8961 -> 0.8980 | Improved |
| add zero initialization (OPRO vs w/o zero init) (ablation) | ICEdit + OPRO · MagicBrush | DINO: 0.8510 -> 0.8531 | Improved |
| add isometry via OPRO instead of additive bias (APB) (ablation) | ICEdit · MagicBrush | L1: 0.0966 -> 0.0781 | Improved |
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.
- InsertAnything — Insert Anything: Image Insertion via In-Context Editing in DiT (AAAI 2026) · better on 1 of 1 shared result
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
- Enabling Instructional Image Editing with In-Context Generation in Large Scale Diffusion Transformer (NeurIPS 2025 · similarity 0.84)
- UNIC-Adapter: Unified Image-instruction Adapter with Multi-modal Transformer for Image Generation (CVPR 2025 · similarity 0.84)
- EditAR: Unified Conditional Generation with Autoregressive Models (CVPR 2025 · similarity 0.84)
- Invertible Consistency Distillation for Text-Guided Image Editing in Around 7 Steps (NeurIPS 2024 · similarity 0.84)
- Control and Realism: Best of Both Worlds in Layout-to-Image without Training (ICML 2025 · similarity 0.84)
- R&B: Region and Boundary Aware Zero-shot Grounded Text-to-image Generation (ICLR 2024 · similarity 0.84)
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.
