Research map / 3D vision / 3D generation
InstructMix2Mix: Consistent Sparse-View Editing Through Multi-View Model Personalization
Daniel Gilo, Or Litany
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 |
|---|---|---|---|
| distill teacher edits into multi-view student instead of student only (ablation) | I-Mix2Mix · Ablation, 6 edits | CLIP Directional Consistency: 0.014 -> 0.337 | Improved |
| add multi-view student to teacher-only per-view editing (ablation) | I-Mix2Mix · Ablation, 6 edits | CLIP Directional Consistency: 0.228 -> 0.337 | Improved |
| edit the reference frame with the teacher before initialization (ablation) | I-Mix2Mix · Ablation, 6 edits | CLIP Directional Consistency: 0.326 -> 0.337 | Improved |
| replace bilinear alignment with a learned conv mapping (ablation) | I-Mix2Mix · Ablation, 6 edits | CLIP Directional Consistency: 0.337 -> 0.287 | No clear effect |
| use stochastic truncated-normal teacher timestep instead of uniform t (ablation) | I-Mix2Mix · Ablation, 6 edits | CLIP Directional Similarity: 0.146 -> 0.178 | Improved |
| use stochastic truncated-normal teacher timestep instead of student-matched t (ablation) | I-Mix2Mix · Ablation, 6 edits | CLIP Directional Similarity: 0.107 -> 0.178 | Improved |
| add Random Cross-View Attention in teacher (ablation) | I-Mix2Mix · Ablation, 6 edits | CLIP Directional Consistency: 0.230 -> 0.337 | Improved |
| add I-Mix2Mix versus DGE | multi-view editing · Sparse-view editing, N=4, 20 edits on 3 I-N2N scenes | CLIP Directional Consistency: 0.287 -> 0.342 | Improved |
| add I-Mix2Mix versus DGE | multi-view editing · Sparse-view editing, N=4, 20 edits on 3 I-N2N scenes | CLIP Directional Similarity: 0.182 -> 0.173 | Hurt |
| replace single-step student prediction with three-step alternative (ablation) | I-Mix2Mix · N=4 peak GPU memory | memory | Hurt |
| use latent interpolation instead of decoder-encoder alignment (ablation) | I-Mix2Mix · peak GPU memory | memory | Improved |
| fine-tune student with LoRA instead of full U-Net (ablation) | I-Mix2Mix · editing | quality | Hurt |
Closest papers by meaning
- ConsistDreamer: 3D-Consistent 2D Diffusion for High-Fidelity Scene Editing (CVPR 2024 · similarity 0.88)
- EditSplat: Multi-View Fusion and Attention-Guided Optimization for View-Consistent 3D Scene Editing with 3D Gaussian Splatting (CVPR 2025 · similarity 0.88)
- SyncNoise: Geometrically Consistent Noise Prediction for Instruction-based 3D Editing (AAAI 2025 · similarity 0.88)
- ViewFusion: Towards Multi-View Consistency via Interpolated Denoising (CVPR 2024 · similarity 0.88)
- InsightEdit: Towards Better Instruction Following for Image Editing (CVPR 2025 · similarity 0.87)
- Instruct-CLIP: Improving Instruction-Guided Image Editing with Automated Data Refinement Using Contrastive Learning (CVPR 2025 · 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.
