Research map / Detection and segmentation / Open-set and few-shot detection
GOOD: Exploring geometric cues for detecting objects in an open world
Haiwen Huang, Andreas Geiger, Dan Zhang
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 depth-based pseudo boxes from geometry to RGB detector training | OLN · COCO person→non-person | AR_N@100: 16.5 -> 25.6 | Improved |
| Add depth- and normal-based pseudo boxes together (ablation) | GOOD on COCO VOC→non-VOC · COCO VOC→non-VOC | AR_N@100: 39.0 -> 39.3 | Improved |
| Use geometric pseudo boxes instead of RGB-only self-training | OLN · COCO VOC→non-VOC | AR_N@100: 37.4 -> 39.3 | Improved |
| Use geometric pseudo boxes instead of RGB-only self-training | OLN · VOC→ADE20K | AR_A@100: 29.2 -> 34.0 | Improved |
| Use geometry rather than edge-map pseudo labels (ablation) | GOOD · COCO VOC→non-VOC | AR_N@100: 38.1 -> 39.0 | Improved |
| Use geometry rather than pairwise-affinity pseudo labels (ablation) | GOOD · COCO VOC→non-VOC | AR_N@100: 37.1 -> 39.0 | Improved |
| Replace the SelfTrain-RGB backbone with stylized ImageNet-pretrained weights (ablation) | SelfTrain-RGB · VOC→non-VOC | AR_N@100: 37.4 -> 36.5 | Hurt |
| Use multi-scale augmentation to collect RGB pseudo boxes (ablation) | SelfTrain-RGB · VOC→non-VOC | AR_N@100: 37.4 -> 37.9 | Improved |
| Add RGB-based pseudo boxes to the depth-and-normal ensemble (ablation) | GOOD-Both · VOC→ADE20K | AR_A@100: 34.0 -> 33.3 | Hurt |
| Train the Phase-II detector with AutoAugment (ablation) | GOOD · COCO VOC→non-VOC | AR_N@100 | Improved |
| Train OLN with AutoAugment on base-class ground-truth boxes (ablation) | OLN · COCO VOC→non-VOC | AR_N@100 | Hurt |
| Stack RGB with geometric cues in Phase I (ablation) | GOOD · Open-world benchmarks | AR_N@100 | Hurt |
Closest papers by meaning
- Towards 3D Objectness Learning in an Open World (NeurIPS 2025 · similarity 0.89)
- Open-World Objectness Modeling Unifies Novel Object Detection (CVPR 2025 · similarity 0.88)
- Exploring Orthogonality in Open World Object Detection (CVPR 2024 · similarity 0.88)
- VL-SAM-V2: Open-World Object Detection with General and Specific Query Fusion (NeurIPS 2025 · similarity 0.87)
- The Devil is in the Fine-Grained Details: Evaluating Open-Vocabulary Object Detectors for Fine-Grained Understanding (CVPR 2024 · similarity 0.87)
- Generative Region-Language Pretraining for Open-Ended Object Detection (CVPR 2024 · 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.
