Research map / Detection and segmentation / 3D detection and driving
Finding Your (3D) Center: 3D Object Detection Using a Learned Loss
David Griffiths, Jan Boehm, Tobias Ritschel
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 |
|---|---|---|---|
| train the scene network with the learned loss on unlabeled scenes instead of sliding the loss network | Ours · S3D chair, 100% labels | mAP@.25: .598 -> .803 | Improved |
| train with the learned loss instead of supervised Chamfer loss | Ours · S3D chair, 5% labels | mAP@.25: .352 -> .642 | Improved |
| train with the learned loss instead of supervised Chamfer loss | Ours · S3D chair, 100% labels | mAP@.25: .756 -> .803 | Improved |
| train with the learned loss instead of sliding window with very few labels | Ours · S3D chair, 1% labels | mAP@.25: .473 -> .366 | Hurt |
| increase label ratio from 1% to 100% | Ours · S3D chair | mAP@.25: .366 -> .803 | Improved |
| increase label ratio from 1% to 100% | Ours · S3D chair | Chamfer error: .554 -> .178 | Improved |
| add unlabeled ScanNet scenes to scene network training (loss net from S3D) | Ours · ScanNet test, 5% S3D labels | mAP@.25: .138 -> .599 | Improved |
| increase number of unlabeled ScanNet training scenes | Ours · ScanNet | mAP@.25: .079 -> .599 | Improved |
| increase loss network labels from 1% to 100% | Loss network · S3D chair | accuracy: .853 -> .960 | Improved |
| use learned loss instead of supervised or sliding on other classes | Ours · S3D bookcase, 5% labels | mAP@.25: .370 -> .640 | Improved |
| reduce labels from 100% to 5% on ScanNet V2 | Ours · ScanNet V2 | mAP@.25 (%): 50.2 -> 48.3 | No clear effect |
| add overlap avoidance (winner-takes-all collision penalty) during scene network training (ablation) | Ours · scene network training | detection | Improved |
Closest papers by meaning
- RevealNet: Seeing Behind Objects in RGB-D Scans (CVPR 2020 · similarity 0.90)
- Omni3D: A Large Benchmark and Model for 3D Object Detection in the Wild (CVPR 2023 · similarity 0.89)
- Uni3D: A Unified Baseline for Multi-Dataset 3D Object Detection (CVPR 2023 · similarity 0.89)
- Point2Seq: Detecting 3D Objects As Sequences (CVPR 2022 · similarity 0.89)
- CenterFormer: Center-based Transformer for 3D Object Detection (ECCV 2022 · similarity 0.89)
- SP3D: Boosting Sparsely-Supervised 3D Object Detection via Accurate Cross-Modal Semantic Prompts (CVPR 2025 · similarity 0.88)
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.
