Research map / Point clouds / Point cloud classification
CLIPoint3D: Language-Grounded Few-Shot Unsupervised 3D Point Cloud Domain Adaptation
Mainak Singha, Sarthak Mehrotra, Paolo Casari, Subhasis Chaudhuri, Elisa Ricci, Biplab Banerjee
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 |
|---|---|---|---|
| apply CLIPoint3D (LoRA on vision encoder + prompt tuning) instead of encoder-based 3D UDA | CLIP ViT-B/16 · PointDA-10 avg | accuracy (%): 72.2 -> 75.7 | Improved |
| adapt CLIP with CLIPoint3D versus best encoder-based baseline | CLIP ViT-B/16 · GraspNetPC-10 avg | accuracy (%): 75.8 -> 92.2 | Improved |
| CLIPoint3D on ScanNet target directions | CLIP ViT-B/16 · PointDA-10 ModelNet->ScanNet | accuracy (%): 57.1 -> 53.5 | Hurt |
| apply LoRA to text encoder only (CLIPoint3D-T) (ablation) | CLIPoint3D · PointDA-10 avg | accuracy (%): 50.7 vs 75.7 (vision) | Hurt |
| add LoRA on vision encoder in addition to prompt tuning (ablation) | CLIPoint3D (GraspNetPC-10) · GraspNetPC-10 avg | accuracy (%): 78.9 -> 90.7 | Improved |
| add knowledge-driven prompt tuning to LoRA on both encoders (ablation) | CLIPoint3D (GraspNetPC-10) · GraspNetPC-10 avg | accuracy (%): 90.5 -> 92.2 | Improved |
| use LoRA instead of LayerNorm tuning or BitFit (ablation) | CLIPoint3D · GraspNetPC-10 avg | accuracy (%): LoRA 90.5 vs BitFit 81.1 vs LayerNorm 76.6 | Improved |
| add OT alignment loss to Lce+Lortho (ablation) | CLIPoint3D · PointDA-10 avg | accuracy (%): 58.6 -> 70.4 | Improved |
| add prototype alignment loss to Lce+Lortho (ablation) | CLIPoint3D · PointDA-10 avg | accuracy (%): 58.6 -> 64.5 | Improved |
| add all alignment losses (proto, OT, conf) to Lce+Lortho (ablation) | CLIPoint3D · GraspNetPC-10 avg | accuracy (%): 74.9 -> 92.2 | Improved |
| use LLM-guided text prompt and 3D-conditioned visual prompt instead of plain learned prompts (ablation) | CLIPoint3D · PointDA-10 avg | accuracy (%): 72.4 -> 75.7 | Improved |
| replace entropy-guided view selection with random single view (ablation) | CLIPoint3D-B · GraspNetPC-10 | accuracy (%): 70.9 -> 92.2 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- PointCLIPv2 — PointCLIP V2: Prompting CLIP and GPT for Powerful 3D Open-world Learning (ICCV 2023) · better on 1 of 1 shared result
- PointDAN — PointDAN: A Multi-Scale 3D Domain Adaption Network for Point Cloud Representation (NeurIPS 2019) · 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
- CLIP2Point: Transfer CLIP to Point Cloud Classification with Image-Depth Pre-Training (ICCV 2023 · similarity 0.92)
- More Text, Less Point: Towards 3D Data-Efficient Point-Language Understanding (AAAI 2025 · similarity 0.89)
- Point2Real: Bridging the Gap between Point Cloud and Realistic Image for Open-World 3D Recognition (AAAI 2024 · similarity 0.89)
- On Geometry-Enhanced Parameter-Efficient Fine-Tuning for 3D Scene Segmentation (NeurIPS 2025 · similarity 0.88)
- VLM-3R: Vision-Language Models Augmented with Instruction-Aligned 3D Reconstruction (CVPR 2026 · similarity 0.88)
- Exploring Vision Semantic Prompt for Efficient Point Cloud Understanding (ICML 2025 · similarity 0.88)
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.
