Research map / Detection and segmentation / Open-set and few-shot detection
Unsupervised Domain Adaptive Detection with Network Stability Analysis
Wenzhang Zhou, Heng Fan, Tiejian Luo, Libo Zhang
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 Network Stability Analysis (HID+LID+InsD) with teacher-student and pseudo-label stage S3 | Faster R-CNN baseline w. data aug. · Cityscapes->FoggyCityscapes | mAP: 34.2 -> 52.7 | Improved |
| add heavy image-level disturbance (HID) consistency (ablation) | NSA-UDA · Cityscapes->FoggyCityscapes (S2) | mAP: 34.2 -> 44.2 | Improved |
| add light image-level disturbance (LID) on top of HID (ablation) | NSA-UDA · Cityscapes->FoggyCityscapes (S2) | mAP: 44.2 -> 48.9 | Improved |
| add instance-level disturbance (InsD) on top of HID and LID (ablation) | NSA-UDA · Cityscapes->FoggyCityscapes (S2) | mAP: 48.9 -> 49.6 | Improved |
| add external consistency analysis (ECA) to NSA_LID (ablation) | NSA-UDA · Cityscapes->FoggyCityscapes (S2) | mAP: 45.9 -> 47.2 | Improved |
| add internal consistency analysis (ICA) to NSA_LID (ablation) | NSA-UDA · Cityscapes->FoggyCityscapes (S2) | mAP: 45.9 -> 48.7 | Improved |
| apply NSA's disturbances to PT instead of its original perturbation (ablation) | PT · Cityscapes->FoggyCityscapes | mAP: 42.7 -> 44.9 | Improved |
| add pseudo-label stage S3 after source-only NSA stage S2 (ablation) | NSA-UDA · KITTI->FoggyCityscapes | mAP: 41.9 -> 50.0 | Improved |
| apply NSA (stage S2) using source data only (ablation) | Faster R-CNN baseline w. data aug. · SIM10k->Cityscapes | AP car: 44.2 -> 52.2 | Improved |
| plug NSA into one-stage FCOS detector | FCOS · Cityscapes->FoggyCityscapes | mAP: 21.0 -> 44.2 | Improved |
| compare NSA-UDA with PT on KITTI->Cityscapes | NSA-UDA · KITTI->Cityscapes | AP car: 60.2 (PT) vs 55.6 | Hurt |
| vary LID loss weight gamma_LID (ablation) | NSA-UDA · Cityscapes->FoggyCityscapes (S2) | mAP: 48.3 / 49.6 / 49.2 / 48.7 over 6e-4 to 6e-1 | Mixed |
Closest papers by meaning
- Domain-Specific Suppression for Adaptive Object Detection (CVPR 2021 · similarity 0.90)
- Domain Adaptive Object Detection via Asymmetric Tri-way Faster-RCNN (ECCV 2020 · similarity 0.89)
- Domain Adaptive Faster R-CNN for Object Detection in the Wild (CVPR 2018 · similarity 0.89)
- Robust Domain Adaptive Object Detection With Unified Multi-Granularity Alignment (TPAMI 2024 · similarity 0.89)
- CAT: Exploiting Inter-Class Dynamics for Domain Adaptive Object Detection (CVPR 2024 · similarity 0.88)
- Universal Domain Adaptive Object Detection via Dual Probabilistic Alignment (AAAI 2025 · 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.
