Research map / Vision-language and multimodal / CLIP and zero-shot
SVL: Empowering Spiking Neural Networks for Efficient 3D Open-World Understanding
Xuerui Qiu, Shaowei Gu, Peixi Wu, JiaKui Hu, Yaozhi Wen, Yuqi Pan, Xinhao Luo, Bo XU, Guoqi Li
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 spike-image NCE alignment to no-alignment baseline (ablation) | SVL pretraining (E-3DSNN-T) · Objaverse-LVIS zero-shot | top-1 accuracy (%): 0.5 -> 24.8 | Improved |
| add spike-text NCE alone (no image alignment) (ablation) | SVL pretraining (E-3DSNN-T) · Objaverse-LVIS zero-shot | top-1 accuracy (%): 0.5 -> 21.9 | Improved |
| combine spike-text NCE with spike-image NCE (ablation) | SVL pretraining (E-3DSNN-T) · Objaverse-LVIS zero-shot | top-1 accuracy (%): 24.8 -> 31.7 | Improved |
| add fine-grained spike-image MSE loss (ablation) | SVL pretraining (E-3DSNN-T) · Objaverse-LVIS zero-shot | top-1 accuracy (%): 31.7 -> 33.6 | Improved |
| add fine-grained spike-image MSE loss (ablation) | SVL pretraining (E-3DSNN-T) · ModelNet40 zero-shot | top-1 accuracy (%): 77.8 -> 79.6 | Improved |
| increase pretraining T from 2 to 4 at D=1 (ablation) | SVL pretraining (E-3DSNN-T) · Objaverse-LVIS zero-shot | top-1 accuracy (%): 32.7 -> 32.9 | No clear effect |
| increase firing bits D from 2 to 4 at T=1 (ablation) | SVL pretraining (E-3DSNN-T) · Objaverse-LVIS zero-shot | top-1 accuracy (%): 32.9 -> 33.6 | Improved |
| use larger OpenCLIP image encoder (ViT-B to ViT-bigG) for prototypes (ablation) | SVL pretraining · ModelNet40 zero-shot | top-1 accuracy (%): 66.8 -> 79.6 | Improved |
| use larger OpenCLIP image encoder (ViT-B to ViT-bigG) for prototypes (ablation) | SVL pretraining · Objaverse-LVIS zero-shot | top-1 accuracy (%): 25.1 -> 33.6 | Improved |
| raise voxel size from 0.01 to 0.02 (0.04 gives 79.1) (ablation) | E-3DSNN-T + SVL · ModelNet40 zero-shot | top-1 accuracy (%): 78.9 -> 79.6 | Mixed |
| raise input points from 10k to 20k (ablation) | E-3DSNN-T + SVL · ModelNet40 zero-shot | top-1 accuracy (%): 79.6 -> 79.8 | No clear effect |
| add SVL pretraining to Spike PointNet (ablation) | Spike PointNet · ScanObjectNN fine-tuned | accuracy (%): 70.0 -> 76.1 | Improved |
Closest papers by meaning
- Efficient 3D Recognition with Event-driven Spike Sparse Convolution (AAAI 2025 · similarity 0.89)
- From Flatland to Space: Teaching Vision-Language Models to Perceive and Reason in 3D (NeurIPS 2025 · similarity 0.87)
- GPT4Scene: Understand 3D Scenes from Videos with Vision-Language Models (ICLR 2026 · similarity 0.87)
- Unifying 2D and 3D Vision-Language Understanding (ICML 2025 · similarity 0.87)
- RegionPLC: Regional Point-Language Contrastive Learning for Open-World 3D Scene Understanding (CVPR 2024 · similarity 0.86)
- Point-to-Spike Residual Learning for Energy-Efficient 3D Point Cloud Classification (AAAI 2024 · similarity 0.86)
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.
