Research map / Point clouds / Segmentation and detection
Image-to-Lidar Self-Supervised Distillation for Autonomous Driving Data
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Distill image features using superpixel-to-superpoint contrastive matching instead of pixel-to-point matching (ablation) | SLidR · nuScenes mini-val semantic segmentation, linear probing | mIoU: 36.6 -> 39.2 | Helped |
| Use dilated rather than strided convolutions in the image backbone, together with superpixels (ablation) | PPKT-style image-to-Lidar distillation · nuScenes mini-val semantic segmentation, linear probing | mIoU: 34.7 -> 36.6 | Helped |
| Use 150 SLIC superpixels per image rather than 100 or 200 (ablation) | SLidR · nuScenes mini-val semantic segmentation, linear probing | mIoU: 37.7 -> 39.2; 36.3 -> 39.2 | Helped |
| Replace SLIC with Felzenszwalb superpixels using tuned parameters (ablation) | SLidR · nuScenes mini-val semantic segmentation, linear probing | mIoU: 39.2 -> 39.2 | No clear effect |
| Use self-supervised MoCov2 rather than fully supervised ImageNet pre-training for ResNet-50 (ablation) | SLidR image teacher · nuScenes mini-val semantic segmentation, linear probing | mIoU: 39.2 -> 39.2 | No clear effect |
| Replace ResNet-50 with ViT-S/16 and DINO pre-training (ablation) | SLidR image teacher · nuScenes mini-val semantic segmentation, linear probing | mIoU: 39.2 -> 39.5 | Helped |
| Pre-train with SLidR rather than PPKT | Sparse residual U-Net · nuScenes validation semantic segmentation, linear probing | mIoU: 36.4 -> 38.8 | Helped |
| Pre-train with SLidR rather than PPKT | Sparse residual U-Net · nuScenes validation semantic segmentation, fine-tuning with 1% labels | mIoU: 37.8 -> 38.3 | Helped |
| Pre-train with SLidR rather than PPKT | Sparse residual U-Net · SemanticKITTI sequence 8 semantic segmentation, fine-tuning with 1% labels | mIoU: 43.9 -> 44.6 | Helped |
| Pre-train with SLidR rather than PPKT | PointRCNN · KITTI 3D object detection, fine-tuning with 5% labels | average mAP: 57.8 -> 57.8 | No clear effect |
| Pre-train with SLidR rather than PPKT | PointRCNN · KITTI 3D object detection, fine-tuning with 10% labels | average mAP: 60.1 -> 61.4 | Helped |
| Pre-train with SLidR rather than PPKT | PointRCNN · KITTI 3D object detection, fine-tuning with 20% labels | average mAP: 61.2 -> 62.4 | Helped |
Closest papers by meaning
- Temporal Consistent 3D LiDAR Representation Learning for Semantic Perception in Autonomous Driving (CVPR 2023 · similarity 0.94)
- VLM2Scene: Self-Supervised Image-Text-LiDAR Learning with Foundation Models for Autonomous Driving Scene Understanding (AAAI 2024 · similarity 0.92)
- Three Pillars Improving Vision Foundation Model Distillation for Lidar (CVPR 2024 · similarity 0.91)
- HiLoTs: High-Low Temporal Sensitive Representation Learning for Semi-Supervised LiDAR Segmentation in Autonomous Driving (CVPR 2025 · similarity 0.91)
- PSA-SSL: Pose and Size-aware Self-Supervised Learning on LiDAR Point Clouds (CVPR 2025 · similarity 0.90)
- BEV-MAE: Bird’s Eye View Masked Autoencoders for Point Cloud Pre-training in Autonomous Driving Scenarios (AAAI 2024 · similarity 0.89)
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.