Research map / Vision-language and multimodal / CLIP and zero-shot
PLA: Language-Driven Open-Vocabulary 3D Scene Understanding
Runyu Ding, Jihan Yang, Chuhui Xue, Wenqing Zhang, Song Bai, Xiaojuan Qi
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 image-bridged point-caption supervision (ablation) | PLA · ScanNet semantic B15/N4 | hIoU: 39.7 -> 65.3 | Improved |
| add binary calibration head (ablation) | LSeg-3D · ScanNet semantic B15/N4 | hIoU: 00.0 -> 39.8 | Improved |
| add entity-level caption supervision (ablation) | PLA with binary calibration · ScanNet semantic B15/N4 | hIoU: 39.8 -> 63.6 | Improved |
| add view-level caption supervision (ablation) | PLA with binary calibration · ScanNet semantic B15/N4 | hIoU: 39.8 -> 61.3 | Improved |
| add scene-level caption supervision to view- and entity-level captions (ablation) | PLA with binary calibration and view/entity captions · ScanNet semantic B15/N4 | hIoU: 65.3 -> 64.6 | Hurt |
| retain only entity phrases in generated captions (ablation) | PLA · ScanNet semantic B15/N4 | hIoU: 65.3 -> 65.7 | Improved |
| use only category label names in captions rather than diverse entity phrases (ablation) | PLA · ScanNet semantic B15/N4 | mIoU novel: 62.4 -> 49.6 | Hurt |
| replace CLIP with BERT as text encoder (ablation) | PLA · ScanNet semantic B15/N4 | mIoU novel: 62.4 -> 55.2 | Hurt |
| replace ViT-GPT2 with OFA for image captioning (ablation) | PLA · ScanNet semantic B15/N4 | hIoU: 65.3 -> 65.6 | Improved |
| add self-training with novel-category names | PLA · ScanNet semantic B15/N4 | hIoU: 65.3 -> 70.3 | Improved |
| add point-caption supervision to a fully supervised model (ablation) | Fully-Sup. · ScanNet semantic B15/N4 | mIoU: 70.62 -> 70.82 | No clear effect |
Closest papers by meaning
- OpenLex3D: A Tiered Benchmark for Open-Vocabulary 3D Scene Representations (NeurIPS 2025 · similarity 0.93)
- Cross-Modal and Uncertainty-Aware Agglomeration for Open-Vocabulary 3D Scene Understanding (CVPR 2025 · similarity 0.92)
- Masked Point-Entity Contrast for Open-Vocabulary 3D Scene Understanding (CVPR 2025 · similarity 0.91)
- RegionPLC: Regional Point-Language Contrastive Learning for Open-World 3D Scene Understanding (CVPR 2024 · similarity 0.90)
- Lexicon3D: Probing Visual Foundation Models for Complex 3D Scene Understanding (NeurIPS 2024 · similarity 0.90)
- Open-Vocabulary 3D Semantic Segmentation with Foundation Models (CVPR 2024 · similarity 0.90)
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.
