Research map / Detection and segmentation / 3D detection and driving
TACO: Task-Aware Contrastive Learning for Joint LiDAR Localization and 3D Object Detection
Leyuan Xing, Huanjia Zhang, Dongyu Pan, Hai Wu, Qiming Xia, Kezheng Xiong, Wen Li, Chenglu Wen, Cheng Wang
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 a detection head (joint multi-task training) (ablation) | LiSA localization · OxfoLD test | position/orientation error: 0.95m, 1.14 -> 0.89m, 1.06 | Improved |
| add ITCL (inter-task contrastive) to joint model (ablation) | TACO · OxfoLD test | position error / vehicle AP: 0.89m -> 0.82m; AP 72.54% -> 75.32% | Improved |
| add IDCL (intra-detection contrastive) to joint model (ablation) | TACO · OxfoLD test | Vehicle AP@0.5: 72.54% -> 77.45% | Improved |
| add ILCL (intra-localization contrastive) to joint model (ablation) | TACO · OxfoLD test | position error: 0.89m -> 0.76m | Improved |
| combine ITCL, IDCL and ILCL (ablation) | TACO · OxfoLD test | Vehicle AP@0.5: 72.54% -> 81.60% | Improved |
| train jointly instead of detection-only (ablation) | TACO · OxfoLD test | Vehicle AP@0.5: 60.01% -> 81.60% | Improved |
| train jointly instead of localization-only (ablation) | TACO · OxfoLD test | position error: 0.95m,1.14 -> 0.72m,0.85 | Improved |
| use TACO instead of Cross-stitch multi-task architecture | TACO · OxfoLD test | position error: 1.97m -> 0.72m | Improved |
| use TACO instead of LiDARFormer | TACO · OxfoLD test | position error / vehicle AP: 0.89m -> 0.72m; 72.54% -> 81.60% | Improved |
| use TACO instead of LiDARFormer | TACO · nuScenes | position error: 1.01m -> 0.92m | Improved |
| use TACO instead of LiDARFormer | TACO · KITTI-360 Seq. 09 | Vehicle AP@0.5: 53.21% -> 53.47% | Improved |
| add more training trajectories (4 vs 1) (ablation) | TACO · OxfoLD test | position error: 4.32m -> 0.72m | Improved |
Closest papers by meaning
- TCC-Det: Temporarily consistent cues for weakly-supervised 3D detection (ECCV 2024 · similarity 0.86)
- Implicit Surface Contrastive Clustering for LiDAR Point Clouds (CVPR 2023 · similarity 0.86)
- An Empirical Study of the Generalization Ability of Lidar 3D Object Detectors to Unseen Domains (CVPR 2024 · similarity 0.85)
- CALICO: Self-Supervised Camera-LiDAR Contrastive Pre-training for BEV Perception (ICLR 2024 · similarity 0.85)
- Fine-grained Image-to-LiDAR Contrastive Distillation with Visual Foundation Models (NeurIPS 2024 · similarity 0.85)
- MOS: Model Synergy for Test-Time Adaptation on LiDAR-Based 3D Object Detection (ICLR 2025 · similarity 0.84)
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.
