Research map / Detection and segmentation / Open-set and few-shot detection
Multi-Source Domain Adaptation for Object Detection
Xingxu Yao, Sicheng Zhao, Pengfei Xu, Jufeng Yang
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 low-level domain discriminator loss (ablation) | DMSN source subnets · BDD100k daytime+night -> dawn/dusk | mAP (%): 28.0 -> 29.9 | Improved |
| add high-level domain discriminator loss (ablation) | DMSN · BDD100k daytime+night -> dawn/dusk | mAP (%): 29.9 -> 31.7 | Improved |
| add pseudo subnet (ablation) | DMSN · BDD100k daytime+night -> dawn/dusk | mAP (%): 31.7 -> 32.9 | Improved |
| add instance-level RPN consistency regularization (ablation) | DMSN · BDD100k daytime+night -> dawn/dusk | mAP (%): 32.9 -> 34.2 | Improved |
| add dynamic weighting of source subnets (ablation) | DMSN · BDD100k daytime+night -> dawn/dusk | mAP (%): 34.2 -> 35.0 | Improved |
| combine Cityscapes and KITTI sources instead of Cityscapes only (source-only) | Faster R-CNN · Cityscapes+KITTI -> BDD100k daytime | AP on car (%): 44.6 -> 43.2 | Hurt |
| combine sources instead of Cityscapes-only DA | SW-ICR-CCR · Cityscapes+KITTI -> BDD100k daytime | AP on car (%): 46.5 -> 43.6 | Hurt |
| use DMSN instead of best source-combined DA | DMSN · BDD100k daytime+night -> dawn/dusk | mAP (%): 30.9 -> 35.0 | Improved |
| use DMSN instead of multi-source MDAN | DMSN · BDD100k daytime+night -> dawn/dusk | mAP (%): 27.6 -> 35.0 | Improved |
| use DMSN instead of M3SDA on cross-camera adaptation | DMSN · Cityscapes+KITTI -> BDD100k daytime | AP on car (%): 44.1 -> 49.2 | Improved |
| combine day and night sources in source-only training instead of daytime only | Faster R-CNN · BDD100k daytime+night -> dawn/dusk | mAP (%): 30.4 -> 28.9 | Hurt |
| apply image-level multi-source alignment (MDAN) instead of single-source DA | MDAN · BDD100k daytime+night -> dawn/dusk | mAP (%): 31.8 -> 27.6 | Hurt |
Closest papers by meaning
- iFAN: Image-Instance Full Alignment Networks for Adaptive Object Detection (AAAI 2020 · similarity 0.91)
- Adapting Object Detectors via Selective Cross-Domain Alignment (CVPR 2019 · similarity 0.91)
- Multi-Adversarial Faster-RCNN for Unrestricted Object Detection (ICCV 2019 · similarity 0.91)
- Diversify and Match: A Domain Adaptive Representation Learning Paradigm for Object Detection (CVPR 2019 · similarity 0.90)
- Domain Adaptive Faster R-CNN for Object Detection in the Wild (CVPR 2018 · similarity 0.90)
- Unified Domain Generalization and Adaptation for Multi-View 3D Object Detection (NeurIPS 2024 · 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.
