Research map / 3D vision / 3D generation
TINKER: Diffusion's Gift to 3D--Multi-View Consistent Editing From Sparse Inputs without Per-Scene Optimization
Canyu Zhao, Xiaoman Li, Tianjian Feng, Zhiyue Zhao, Hao Chen, Chunhua Shen
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 |
|---|---|---|---|
| fine-tune FLUX Kontext on referring multi-view editing data (ablation) | FLUX Kontext · Multi-view editing (Mip-NeRF 360, FLUX Kontext) | DINO similarity: 0.862 -> 0.943 | Improved |
| fine-tune FLUX Kontext on referring multi-view editing data (ablation) | FLUX Kontext · Multi-view editing (Mip-NeRF 360, FLUX Kontext) | Aesthetic score: 7.058 -> 6.973 | No clear effect |
| condition on depth instead of ray maps (ablation) | scene completion model · Scene completion conditioning | DINO similarity: 0.931 -> 0.978 | Improved |
| condition on depth instead of ray maps (ablation) | scene completion model · Scene completion conditioning | Aesthetic score: 6.214 -> 6.586 | Improved |
| concatenate more than two images in editing (ablation) | multi-view editing model · Aesthetic score vs number of concatenated images | Aesthetic score | Hurt |
| few-shot instead of one-shot input | TINKER · 3D editing (Mip-NeRF-360 + IN2N) | CLIP-dir: 0.143 -> 0.157 | Improved |
| TINKER vs strongest baseline GaussCtrl | TINKER few-shot · 3D editing (Mip-NeRF-360 + IN2N) | CLIP-dir: 0.123 -> 0.157 | Improved |
| TINKER vs DGE on editing time | TINKER · 3D editing (Mip-NeRF-360 + IN2N) | average editing time: 10min -> 15min | Hurt |
| replace VACE with TINKER depth-conditioned model | scene completion · Video reconstruction | PSNR: 16.635 -> 31.869 | Improved |
| add mask to VACE | VACE · Scene completion conditioning | DINO similarity: 0.916 -> 0.954 | 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.
- EditSplat — EditSplat: Multi-View Fusion and Attention-Guided Optimization for View-Consistent 3D Scene Editing with 3D Gaussian Splatting (CVPR 2025) · better on 3 of 3 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
- Coupled Diffusion Sampling for Training-Free Multi-View Image Editing (CVPR 2026 · similarity 0.91)
- EditSplat: Multi-View Fusion and Attention-Guided Optimization for View-Consistent 3D Scene Editing with 3D Gaussian Splatting (CVPR 2025 · similarity 0.90)
- ProEdit: Simple Progression is All You Need for High-Quality 3D Scene Editing (NeurIPS 2024 · similarity 0.89)
- SyncNoise: Geometrically Consistent Noise Prediction for Instruction-based 3D Editing (AAAI 2025 · similarity 0.88)
- ConsistDreamer: 3D-Consistent 2D Diffusion for High-Fidelity Scene Editing (CVPR 2024 · similarity 0.88)
- Instruct 4D-to-4D: Editing 4D Scenes as Pseudo-3D Scenes Using 2D Diffusion (CVPR 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.
