Research map / Detection and segmentation / Open-set and few-shot detection
Laser: Efficient Language-Guided Segmentation in Neural Radiance Fields
Xingyu Miao, Haoran Duan, Yang Bai, Tejal Shah, Jun Song, Yang Long, Rajiv Kumar Ranjan, Ling Shao
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 adapter layer to dense CLIP distillation (ablation) | Laser · 3D-OVS (overall, 6 scenes) | mIoU: 59.4 -> 68.2 | Improved |
| add self-cross-training strategy on top of adapter (ablation) | Laser · 3D-OVS (overall, 6 scenes) | mIoU: 68.2 -> 73.9 | Improved |
| add low-rank transient query attention (ablation) | Laser · 3D-OVS (overall, 6 scenes) | mIoU: 73.9 -> 78.6 | Improved |
| add label volume classification (ablation) | Laser · 3D-OVS (overall, 6 scenes) | mIoU: 78.6 -> 85.6 | Improved |
| add simplified text augmentation loss (ablation) | Laser · 3D-OVS (overall, 6 scenes) | mIoU: 85.6 -> 88.1 | Improved |
| use only reconstructed features (alpha 1.0) instead of alpha 0.2 (ablation) | Laser · 3D-OVS (overall, 6 scenes) | mIoU: 88.1 -> 71.8 | Hurt |
| use original CLIP features only (alpha 0) instead of alpha 0.2 (ablation) | Laser · 3D-OVS (overall, 6 scenes) | mIoU: 88.1 -> 81.8 | Hurt |
| set beta 0.3 instead of 0 in self-cross-training loss (ablation) | Laser · 3D-OVS (overall, 6 scenes) | mIoU: 87.2 -> 88.1 | Improved |
| set gamma 0.5 instead of 0.1 in label volume ensemble (ablation) | Laser · 3D-OVS (overall, 6 scenes) | mIoU: 84.1 -> 88.1 | Improved |
| add transient query attention versus no attention block (ablation) | Laser · 3D-OVS (overall, 6 scenes) | mIoU: 85.31 -> 88.10 | Improved |
| use transient query attention instead of vanilla self-attention (ablation) | Laser · 3D-OVS (overall, 6 scenes) | mIoU: 87.98 -> 88.10 | No clear effect |
| use transient query attention instead of vanilla self-attention (ablation) | Laser · 3D-OVS (overall, 6 scenes) | training time: 2.2 hrs -> 11 min | Improved |
Closest papers by meaning
- FastLGS: Speeding Up Language Embedded Gaussians with Feature Grid Mapping (AAAI 2025 · similarity 0.85)
- Building Vision-Language Models on Solid Foundations with Masked Distillation (CVPR 2024 · similarity 0.85)
- LESS: Label-Efficient and Single-Stage Referring 3D Segmentation (NeurIPS 2024 · similarity 0.85)
- COS3D: Collaborative Open-Vocabulary 3D Segmentation (NeurIPS 2025 · similarity 0.85)
- Rethinking Open-Vocabulary Segmentation of Radiance Fields in 3D Space (AAAI 2025 · similarity 0.85)
- Segment Any 3D Object with Language (ICLR 2025 · similarity 0.85)
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.
