Research map / Detection and segmentation / Open-set and few-shot detection
Improving the Intra-Class Long-Tail in 3D Detection via Rare Example Mining
Chiyu Max Jiang, Mahyar Najibi, Charles R. Qi, Yin Zhou, Dragomir Anguelov
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 |
|---|---|---|---|
| raise mining budget from 3% to 6% of tracks (ablation) | MD-REM++ · Waymo vehicles (Large) | AP@0.5: 0.571 -> 0.612 | Improved |
| raise mining budget from 6% to 9% of tracks (ablation) | MD-REM++ · Waymo vehicles (Large) | AP@0.5: 0.612 -> 0.606 | No clear effect |
| raise mining budget from 3% to 9% of tracks (ablation) | MD-REM++ · Waymo vehicles (All) | AP@0.5: 0.904 -> 0.905 | No clear effect |
| add hard-example filter to data-centric REM (MD-REM++ vs D-REM++) (ablation) | REM · Waymo vehicles (Large) | AP@0.5: 0.533 -> 0.571 | Improved |
| add a strong offboard auto-labeler (D-REM to D-REM++) (ablation) | D-REM · Waymo vehicles (All) | AP@0.5: 0.882 -> 0.906 | Improved |
| add hard example filtering (model-centric REM) over ensemble variance mining (ablation) | Ensemble mining · Mined tracks composition | ratio of large objects: 13.72% -> 24.61% | Improved |
| use data-centric REM instead of model-centric REM for mining (ablation) | REM · Mined tracks composition | ratio of large objects: 24.61% -> 30.60% | Improved |
| combine model and data-centric REM (ablation) | REM · Mined tracks composition | ratio of large objects: 30.60% -> 31.86% | Improved |
| mine by oracle error (Oracle Hard) rather than random (ablation) | base MVF model · Waymo vehicles (Large) | AP@0.5: 0.355 -> 0.341 | No clear effect |
| mine by oracle size (Oracle Size) (ablation) | base MVF model · Waymo vehicles (Regular) | AP@0.5: 0.869 -> 0.583 (All vs Large) | Mixed |
| mine by predicted size heuristic instead of random (ablation) | base MVF model · Waymo vehicles | AP@0.5: 0.355 -> 0.498 (Large), 0.873 -> 0.865 (All) | Mixed |
| add REM-mined labeled tracks (10%;3%) to partial supervision (ablation) | MVF · Waymo vehicles (Large) | AP@0.5: 0.378 -> 0.571 | Improved |
Closest papers by meaning
- Class-Difficulty Based Methods for Long-Tailed Visual Recognition (IJCV 2022 · similarity 0.88)
- Hierarchical Novelty Detection for Visual Object Recognition (CVPR 2018 · similarity 0.87)
- Overcoming Classifier Imbalance for Long-Tail Object Detection With Balanced Group Softmax (CVPR 2020 · similarity 0.87)
- Progress and Limitations of Deep Networks to Recognize Objects in Unusual Poses (AAAI 2023 · similarity 0.87)
- NBA3D: Neighbor-Based Confidence Adjustment for 3D Rare Object Detection Using LiDAR (AAAI 2025 · similarity 0.86)
- Uncertainty Meets Diversity: A Comprehensive Active Learning Framework for Indoor 3D Object Detection (CVPR 2025 · 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.
