Research map / Detection and segmentation / 3D detection and driving
Self-Supervised Sparse Sensor Fusion for Long Range Perception
Edoardo Palladin, Samuel Brucker, Filippo Ghilotti, Praveen Narayanan, Mario Bijelic, Felix Heide
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 occupancy-velocity self-supervised pre-training (ablation) | LRS4Fusion · Long Range Dataset 3D detection | mAP: 49.58 -> 52.61 | Improved |
| add velocity decoder to pre-training (ablation) | LRS4Fusion · Long Range Dataset LiDAR forecasting (1s history) | Chamfer Distance (1s): 16.592 -> 15.821 | Improved |
| add velocity decoder to pre-training (ablation) | LRS4Fusion · Long Range Dataset LiDAR forecasting (1s history) | Chamfer Distance (3s): 42.152 -> 39.031 | Improved |
| use Vim backbone instead of ResNet50 for detection (ablation) | LRS4Fusion · Long Range Dataset 3D detection | mAP: 50.22 -> 52.61 | Improved |
| use 1s history instead of 3s history (ablation) | LRS4Fusion · Long Range Dataset 3D detection | mAP: 51.86 -> 52.61 | Improved |
| add pre-training with 3s history (ablation) | LRS4Fusion · Long Range Dataset 3D detection | mAP: 51.16 -> 51.86 | Improved |
| add temporal history (1s) versus none, stage 1 (ablation) | LRS4Fusion · Long Range Dataset 3D detection | mAP: 50.75 -> 49.58 | Hurt |
| replace Vim with Resnet50 depth backbone (reverse: use Vim) (ablation) | depth model · Long-range depth | memory: 2.9GB -> 1.3GB | Improved |
| add DINOv2 distillation to depth model (ablation) | depth model · Long-range depth | MAE: 3.83 -> 3.61 | Improved |
| replace GRU with MGU update block (ablation) | depth model · Long-range depth | runtime: 245ms -> 213ms | Improved |
| replace OGNI-DC integration with proposed depth integration module (ablation) | depth model · Long-range depth | runtime: 213ms -> 63ms | Improved |
| add camera tokens to image encoding (ablation) | depth model · Long-range depth | MAE: 3.51 -> 3.46 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- BEVFormer — BEVFormer: Learning Bird’s-Eye-View Representation from Multi-Camera Images via Spatiotemporal Transformers (ECCV 2022) · better on 1 of 1 shared result
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
- Fully Sparse 3D Object Detection (NeurIPS 2022 · similarity 0.90)
- TruckDrive: Long-Range Autonomous Highway Driving Dataset (CVPR 2026 · similarity 0.90)
- One Million Scenes for Autonomous Driving: ONCE Dataset (NeurIPS 2021 · similarity 0.89)
- Multi-Modal 3D Object Detection in Autonomous Driving: A Survey (IJCV 2023 · similarity 0.89)
- BEV-MAE: Bird’s Eye View Masked Autoencoders for Point Cloud Pre-training in Autonomous Driving Scenarios (AAAI 2024 · similarity 0.87)
- Effective Multi-sensor Conditioning for Street-view Novel-view Synthesis (NeurIPS 2026 · 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.
