Research map / 3D vision / 3D generation
TASE: Truncation-Aware Semantic Embeddings for 3D Scene Understanding and Editing
Tim-Felix Faasch, Jochen Kall, Jens Behley, Lucas Nunes, Cyrill Stachniss
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 equivariance/consistency loss to the TASE embedding (ablation) | TASE · Feature reconstruction (held-out views) | PSNR: 23.89 -> 24.64 | Improved |
| add equivariance/consistency loss to the TASE embedding (ablation) | TASE · Feature reconstruction (held-out views) | feature loss L_f: 0.1016 -> 0.0120 | Improved |
| add equivariance/consistency loss to the TASE embedding (ablation) | TASE · Local 3D scene editing | CLIP directional similarity: 0.0852 -> 0.1385 | Improved |
| add Difix3D+ finetuning of the editing diffusion model (ablation) | TASE · Local 3D scene editing | CLIP directional similarity: 0.1385 -> 0.1542 | Improved |
| add Difix3D+ finetuning to M_TASE without consistency loss (ablation) | TASE · Local 3D scene editing | CLIP directional similarity: 0.0852 -> 0.0905 | Improved |
| add Difix3D+ finetuning to PCA embedding (ablation) | PCA baseline · Local 3D scene editing | CLIP directional similarity: 0.1067 -> 0.1365 | Improved |
| use autoencoder embedding with consistency loss instead of PCA (ablation) | TASE · Local 3D scene editing | CLIP directional similarity: 0.1365 -> 0.1542 | Improved |
| use autoencoder embedding without consistency loss instead of PCA (ablation) | TASE · Feature reconstruction (held-out views) | feature loss L_f: 0.0879 -> 0.1016 | Hurt |
| use autoencoder embedding without consistency loss instead of PCA (ablation) | TASE · Local 3D scene editing | CLIP directional similarity: 0.1067 -> 0.0852 | Hurt |
| use TASE editing instead of DGE | TASE · Local 3D scene editing | user preference (geometry): 19.1% -> 64.3% | Improved |
| use TASE editing instead of DGE on face scene | TASE · Face local edits | CLIP directional similarity: 0.1181 -> 0.0965 | Hurt |
| retain fewer channels of the truncated embedding (2 vs 64) (ablation) | TASE · 2D image generation with prompt 'in the snow' | qualitative prompt adherence: 2 dims vs 64 dims | 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 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
- Edit3r: Instant 3D Scene Editing from Sparse Unposed Images (ECCV 2026 · similarity 0.87)
- DreamEdit3D: Personalization of Multi-View Diffusion Models for 3D Editing (ECCV 2026 · similarity 0.86)
- OBJECT 3DIT: Language-guided 3D-aware Image Editing (NeurIPS 2023 · similarity 0.86)
- Masked Point-Entity Contrast for Open-Vocabulary 3D Scene Understanding (CVPR 2025 · similarity 0.84)
- Cross-Modal and Uncertainty-Aware Agglomeration for Open-Vocabulary 3D Scene Understanding (CVPR 2025 · similarity 0.84)
- Merge3D: Efficient 3D Multimodal LLMs via Joint 2D-3D Token Merging (CVPR 2026 · 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.
