Research map / Vision-language and multimodal / CLIP and zero-shot
Copy-Transform-Paste: Zero-Shot Object-Object Alignment Guided by Vision-Language and Geometric Constraints
Rotem Gatenyo, Ohad Fried
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 CLIP text guidance (ablation) | Copy-Transform-Paste · Object-object alignment benchmark | CLIP score: 0.3091 -> 0.3224 | Improved |
| add fractional soft-ICP attachment loss (ablation) | Copy-Transform-Paste · Object-object alignment benchmark | Text-Asset Alignment (GPT-4V): 969.40 -> 1028.72; intersection 0.0002 -> 0.0112 | Mixed |
| add penetration loss (ablation) | Copy-Transform-Paste · Object-object alignment benchmark | Intersection volume: 0.0177 -> 0.0112 | Improved |
| add phased optimization with increasing weights (ablation) | Copy-Transform-Paste · Object-object alignment benchmark | Intersection volume: 0.0141 -> 0.0112 | Improved |
| add camera scheduling (ablation) | Copy-Transform-Paste · Object-object alignment benchmark | ALIGN score: 14.5859 -> 14.9800 | Improved |
| use CLIP guidance instead of score distillation sampling (ablation) | Copy-Transform-Paste · Object-object alignment benchmark | CLIP score: 0.3167 -> 0.3224 | Improved |
| use CLIP instead of SigLIP as guidance model (ablation) | Copy-Transform-Paste · Object-object alignment benchmark | SigLIP score: SigLIP guidance 0.0387 vs 0.0380 SigLIP; CLIP 0.3145 vs 0.3224 | Mixed |
| also optimize isotropic scale (ablation) | Copy-Transform-Paste · Object-object alignment benchmark | GPT-4V Overall: 1034.44 -> 1005.10 | Hurt |
| multi-start with CLIP selection for shrinkwrap baseline (ablation) | Shrinkwrap · Object-object alignment benchmark | CLIP score: 0.3087 -> 0.3157 | Improved |
| use LLM to set penetration policy, initial scale and attachment ratio | Copy-Transform-Paste · 61 held-out object pairs | Initial scale MAE: 3.30 -> 0.46 | Improved |
Closest papers by meaning
- A3D: Does Diffusion Dream about 3D Alignment? (ICLR 2025 · similarity 0.88)
- InteractAnything: Zero-shot Human Object Interaction Synthesis via LLM Feedback and Object Affordance Parsing (CVPR 2025 · similarity 0.87)
- SceneForge: Enhancing 3D-text alignment with Structured Scene Compositions (NeurIPS 2025 · similarity 0.87)
- Tango3D: Towards Alignment for Global and Local 2D-3D Correspondence (NeurIPS 2026 · similarity 0.86)
- Object-centric binding in Contrastive Language-Image Pretraining (NeurIPS 2025 · similarity 0.86)
- SweetDreamer: Aligning Geometric Priors in 2D diffusion for Consistent Text-to-3D (ICLR 2024 · 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.
