Research map / Detection and segmentation / Open-set and few-shot detection
BugNIST - a Large Volumetric Dataset for Detection under Domain Shift
Patrick M Jensen, Vedrana A Dahl, Rebecca Engberg, Carsten Gundlach, Hans Martin Kjer, Anders B Dahl
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 on synthetic mixtures instead of single bugs (domain-gap mitigation) (ablation) | Faster R-CNN · BugNIST real mixtures (without class info) | F1-score: 0.56 -> 0.68 | Improved |
| train on synthetic mixtures instead of single bugs (ablation) | nnDetection · BugNIST real mixtures (without class info) | F1-score: 0.03 -> 0.50 | Improved |
| train on crowded synthetic mixtures instead of synthetic mixtures (ablation) | Faster R-CNN · BugNIST real mixtures (without class info) | F1-score: 0.68 -> 0.27 | Hurt |
| train on crowded synthetic mixtures instead of single bugs (ablation) | U-Net · BugNIST real mixtures (without class info) | F1-score: 0.59 -> 0.54 | No clear effect |
| require correct bug class in addition to detection on real mixtures | U-Net · BugNIST real mixtures | F1-score: 0.59 -> 0.11 | Hurt |
| classify individual bugs in-domain (no context shift) | DenseNet · BugNIST individual bug classification | accuracy: 0.11 -> 0.98 | Improved |
| increase synthetic mixes from 300 to 1000 for Faster R-CNN (ablation) | Faster R-CNN · BugNIST training task | detection performance | No clear effect |
| train Faster R-CNN on synthetic mixtures instead of single bugs (in-domain validation) (ablation) | Faster R-CNN · BugNIST training task (validation) | mAP: 0.70 -> 0.02 | Hurt |
| train nnDetection on synthetic mixtures instead of single bugs (in-domain validation) (ablation) | nnDetection · BugNIST training task (validation) | mAP: 0.10 -> 0.59 | Improved |
| larger transformer or ConvNext models for classification vs DenseNet and ResNet | BugNIST classification · BugNIST individual bug classification (x128) | accuracy: 0.97 (DenseNet169) vs 0.87 (ViT) | Hurt |
Closest papers by meaning
- Unsupervised Domain Adaptive 3D Detection With Multi-Level Consistency (ICCV 2021 · similarity 0.83)
- A Dataset for Semantic Segmentation in the Presence of Unknowns (CVPR 2025 · similarity 0.82)
- AEROBLADE: Training-Free Detection of Latent Diffusion Images Using Autoencoder Reconstruction Error (CVPR 2024 · similarity 0.82)
- ImOV3D: Learning Open Vocabulary Point Clouds 3D Object Detection from Only 2D Images (NeurIPS 2024 · similarity 0.82)
- D^3: Scaling Up Deepfake Detection by Learning from Discrepancy (CVPR 2025 · similarity 0.82)
- COUNTS: Benchmarking Object Detectors and Multimodal Large Language Models under Distribution Shifts (CVPR 2025 · similarity 0.81)
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.
