Research map / Detection and segmentation / 3D detection and driving
LIFT: Learning 4D LiDAR Image Fusion Transformer for 3D Object Detection
Yihan Zeng, Da Zhang, Chunwei Wang, Zhenwei Miao, Ting Liu, Xin Zhan, Dayang Hao, Chao Ma
Check 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 |
|---|---|---|---|
| Fuse LiDAR, camera images, and sequential information with sensor-time attention rather than use LiDAR alone (ablation) | PointPillars · nuScenes validation subset (1/8 training) | mAP: 24.83 -> 47.04 | Improved |
| Use self-attention rather than concatenation to fuse LiDAR and images (ablation) | PointPillars · nuScenes validation subset (1/8 training), LiDAR + images | mAP: 41.59 -> 43.22 | Improved |
| Use self-attention rather than concatenation to fuse LiDAR and sequential input (ablation) | PointPillars · nuScenes validation subset (1/8 training), LiDAR + time | mAP: 26.74 -> 27.75 | Improved |
| Increase sequence length from T=2 to T=5 with LIFT attention (ablation) | LIFT · nuScenes validation subset (1/8 training) | mAP: T=2: 27.75; T=5: 30.97 | Improved |
| Replace attention temporal fusion with point concatenation (ablation) | LIFT · nuScenes validation subset (1/8 training), T=2 (+Img) | mAP: 47.04 -> 43.74 | Hurt |
| Add cross-sensor and cross-time-consistent augmentation (ablation) | LIFT · nuScenes validation subset (1/8 training), sequential point-cloud and image input | mAP: 47.04 -> 51.78 | Improved |
| Add point-wise attention within pillars (ablation) | LIFT · nuScenes validation subset (1/8 training) | mAP: 49.76 -> 50.25 | Improved |
| Add 4D relative positional encoding (ablation) | LIFT · nuScenes validation subset (1/8 training) | mAP: 50.25 -> 50.50 | Improved |
| Add pyramid context (ablation) | LIFT · nuScenes validation subset (1/8 training) | mAP: 50.50 -> 51.30 | Improved |
| Add sparse window partition (ablation) | LIFT · nuScenes validation subset (1/8 training) | mAP: 51.30 -> 51.78 | Improved |
| Increase image resolution from 448 × 800 to 896 × 1600 (ablation) | LIFT · nuScenes validation | mAP: 51.78 -> 51.83 | No clear effect |
| Reduce image resolution from 448 × 800 to 224 × 400 (ablation) | LIFT · nuScenes validation | mAP: 51.78 -> 44.20 | Hurt |
Later papers that compared with this method
5 later accepted papers report LiFT as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- UPLiFT: Efficient Pixel-Dense Feature Upsampling with Local Attenders (CVPR 2026 · better on 4 of 4 shared results)
- FlatLab: A Unified Methodology Framework and Simulation-Based Benchmark for Robotic Manipulation of Flat Objects (ICML 2026 · better on 2 of 2 shared results)
- LoftUp: Learning a Coordinate-Based Feature Upsampler for Vision Foundation Models (ICCV 2025 · better on 5 of 5 shared results)
- Adiabatic Quantum Computing for Multi Object Tracking (CVPR 2022 · better on 0 of 2 shared results)
- Learning of Global Objective for Network Flow in Multi-Object Tracking (CVPR 2022 · better on 0 of 2 shared results)
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
- TransFusion: Robust LiDAR-Camera Fusion for 3D Object Detection With Transformers (CVPR 2022 · similarity 0.87)
- DeepFusion: Lidar-Camera Deep Fusion for Multi-Modal 3D Object Detection (CVPR 2022 · similarity 0.87)
- Adaptive-Smooth LiDAR-Camera Knowledge Distillation with Heterogeneous Fusion for Multi-View 3D Object Detection (AAAI 2026 · similarity 0.87)
- MGAF: LiDAR-Camera 3D Object Detection With Multiple Guidance and Adaptive Fusion (TPAMI 2025 · similarity 0.86)
- Self-Supervised Learning of LiDAR 3D Point Clouds via 2D-3D Neural Calibration (TPAMI 2025 · similarity 0.86)
- Towards Universal LiDAR-Based 3D Object Detection by Multi-Domain Knowledge Transfer (ICCV 2023 · 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.
