Research map / 3D vision / 3D generation
CoCoNets: Continuous Contrastive 3D Scene Representations
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 |
|---|---|---|---|
| Use amodal completed point features rather than only visible points | 3DPointContrast-style tracking · CARLA and KITTI object tracking | mean IOU: CARLA: 0.55 -> 0.61; KITTI: 0.48 -> 0.54 | Helped |
| Use continuous high-resolution features rather than discrete voxel features | 3D feature tracking · CARLA and KITTI object tracking | mean IOU: CARLA: 0.37 -> 0.61; KITTI: 0.23 -> 0.54 | Helped |
| Learn features in metric 3D rather than 2D features lifted using depth | 3D object tracking · CARLA and KITTI object tracking | mean IOU: CARLA: 0.24 -> 0.61; KITTI: 0.19 -> 0.54 | Helped |
| Initialize VoteNet point features with CoCoNets pre-training | VoteNet · CARLA 3D object detection | mAP: IoU 0.25: 0.32 -> 0.51; IoU 0.30: 0.26 -> 0.47; IoU 0.40: 0.24 -> 0.41; IoU 0.50: 0.20 -> 0.32 | Helped |
| Use CoCoNets features instead of 3DVoxContrast features | ShapeNet cross-view alignment · ShapeNet cross-view alignment | accuracy: 0.21 -> 0.58 | Helped |
| Use CoCoNets features instead of 3DPointContrast features | ShapeNet cross-view alignment · ShapeNet cross-view alignment | accuracy: 0.14 -> 0.58 | Helped |
| Use CoCoNets instead of 3DPointContrast for cross-object alignment | ShapeNet cross-object alignment · ShapeNet cross-object alignment | accuracy: 0.09 -> 0.18 | Helped |
| Use CoCoNets instead of 3DVoxContrast for cross-object alignment | ShapeNet cross-object alignment · ShapeNet cross-object alignment | accuracy: 0.18 -> 0.18 | No clear effect |
| Use RGB input instead of RGB-D input (ablation) | CoCoNets · ShapeNet cross-object alignment | accuracy: 0.18 -> 0.16 | Hurt |
| Use RGB input instead of RGB-D input (ablation) | CoCoNets · ShapeNet cross-view alignment | accuracy: 0.58 -> 0.39 | Hurt |
| Jointly train RGB and occupancy prediction with contrastive view prediction (ablation) | CoCoNets feature learning · 3D feature discriminability | feature discriminability | No clear effect |
Closest papers by meaning
- Learning from Unlabelled Videos Using Contrastive Predictive Neural 3D Mapping (ICLR 2020 · similarity 0.92)
- Object-Centric Representation Learning for Enhanced 3D Semantic Scene Graph Prediction (NeurIPS 2025 · similarity 0.88)
- OSN: Infinite Representations of Dynamic 3D Scenes from Monocular Videos (ICML 2024 · similarity 0.88)
- Towards Generalizable Multi-Camera 3D Object Detection via Perspective Rendering (AAAI 2025 · similarity 0.87)
- ImageNet3D: Towards General-Purpose Object-Level 3D Understanding (NeurIPS 2024 · similarity 0.87)
- Masked Scene Modeling: Narrowing the Gap Between Supervised and Self-Supervised Learning in 3D Scene Understanding (CVPR 2025 · similarity 0.87)
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.