Research map / Detection and segmentation / Open-set and few-shot detection
R4D: Utilizing Reference Objects for Long-Range Distance Estimation
Yingwei Li, Tiffany Chen, Maya Kabkab, Ruichi Yu, Longlong Jing, Yurong You, Hang Zhao
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 |
|---|---|---|---|
| Use reference-object relationships with attention-based aggregation | R4D · Waymo Open Dataset - Long-Range Labels validation | <10% relative distance error: 53.4% -> 62.3% | Improved |
| Use reference-object relationships with attention-based aggregation | R4D · Pseudo long-range KITTI validation | <10% relative distance error: 65.8% -> 72.5% | Improved |
| Use reference-object relationships with attention-based aggregation | R4D · Pseudo long-range nuScenes validation | <10% relative distance error: 66.7% -> 71.1% | Improved |
| Add attention-weighted aggregation instead of averaging references (ablation) | R4D · Waymo Open Dataset - Long-Range Labels validation | <5% relative distance error: 35.9% -> 38.5%; +2.6% | Improved |
| Add reference embeddings to target-only embeddings (ablation) | R4D · Waymo Open Dataset - Long-Range Labels validation | <10% relative distance error: 53.4% -> 60.8% | Improved |
| Add geo-distance embeddings (ablation) | R4D with target and reference embeddings · Waymo Open Dataset - Long-Range Labels validation | <5% relative distance error: 31.5% -> 36.9% | Improved |
| Add union embeddings alongside geo-distance embeddings (ablation) | R4D · Waymo Open Dataset - Long-Range Labels validation | <5% relative distance error: 36.9% -> 38.5%; +1.6% | Improved |
| Add union embeddings without geo-distance embeddings (ablation) | R4D · Waymo Open Dataset - Long-Range Labels validation | <5% relative distance error: 31.5% -> 32.5% | Hurt |
| Add relative-distance prediction supervision (ablation) | R4D with one reference · Waymo Open Dataset - Long-Range Labels validation, images with references | <10% relative distance error: 55.6% -> 58.8% | Improved |
| Combine absolute- and relative-distance prediction heads (ablation) | R4D-One-Ref · Waymo Open Dataset - Long-Range Labels validation, images with references | <10% relative distance error: 55.6% -> 59.9%; +4.3% | Improved |
| Aggregate multiple references with attention instead of using one reference (ablation) | R4D · Waymo Open Dataset - Long-Range Labels validation, images with references | <10% relative distance error: 59.9% -> 63.3% | Improved |
| Add distance augmentation (ablation) | R4D with multiple references and both distance heads · Waymo Open Dataset - Long-Range Labels validation, images with references | <10% relative distance error: 54.1% -> 63.3% | Improved |
Closest papers by meaning
- Learning Object-Specific Distance From a Monocular Image (ICCV 2019 · similarity 0.88)
- One Million Scenes for Autonomous Driving: ONCE Dataset (NeurIPS 2021 · similarity 0.86)
- DrivingStereo: A Large-Scale Dataset for Stereo Matching in Autonomous Driving Scenarios (CVPR 2019 · similarity 0.86)
- OpenAD: Open-World Autonomous Driving Benchmark for 3D Object Detection (NeurIPS 2025 · similarity 0.84)
- Beyond Chamfer Distance: Granular Order-aware Evaluation Metric For Online Mapping (NeurIPS 2026 · similarity 0.83)
- MAN TruckScenes: A multimodal dataset for autonomous trucking in diverse conditions (NeurIPS 2024 · similarity 0.83)
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.
