Research map / Detection and segmentation / Open-set and few-shot detection
Instance-Invariant Domain Adaptive Object Detection via Progressive Disentanglement
Aming Wu, Yahong Han, Linchao Zhu, Yi 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 |
|---|---|---|---|
| use three-stage training instead of first stage only (ablation) | progressive disentanglement · Cityscapes to FoggyCityscapes | mAP (%): 33.6% to 36.6% | Improved |
| use three-stage training instead of one-stage training of all losses (ablation) | progressive disentanglement · Cityscapes to FoggyCityscapes | mAP (%): 34.1% -> 36.6% | Improved |
| use three-stage training instead of one-stage training (ablation) | progressive disentanglement · Pascal VOC to Watercolor | mAP (%): 52.9% -> 56.9% | Improved |
| add relation-consistency loss (ablation) | progressive disentanglement · Pascal VOC to Watercolor | mAP (%): 55.2% to 56.9% | Improved |
| add relation-consistency loss (ablation) | progressive disentanglement · Cityscapes to FoggyCityscapes | mAP (%): 35.5% -> 36.6% | Improved |
| add third training stage (feature reconstruction) (ablation) | progressive disentanglement · Cityscapes to FoggyCityscapes | mAP (%): 35.3% -> 36.6% | Improved |
| add first disentangled layer (two layers instead of one) (ablation) | progressive disentanglement · Cityscapes to FoggyCityscapes | mAP (%): 34.1% to 36.6% | Improved |
| add first disentangled layer (two layers instead of one) (ablation) | progressive disentanglement · Pascal VOC to Watercolor | mAP (%): 54.6% -> 56.9% | Improved |
| replace SW baseline with instance-invariant progressive disentanglement | SW · Cityscapes to FoggyCityscapes | mAP (%): 34.3 -> 36.57 | Improved |
| replace SW baseline with proposed method | SW · Pascal VOC to Watercolor | mAP (%): 53.3 -> 56.9 | Improved |
| replace SW baseline with proposed method | SW · Pascal VOC to Clipart | mAP (%): 38.1 -> 42.1 | Improved |
| use ResNet101 instead of VGG16 backbone | Ours · Cityscapes to FoggyCityscapes | mAP (%): 36.57 -> 38.63 | Improved |
Closest papers by meaning
- Vector-Decomposed Disentanglement for Domain-Invariant Object Detection (ICCV 2021 · similarity 0.94)
- Domain-Invariant Disentangled Network for Generalizable Object Detection (ICCV 2021 · similarity 0.93)
- Triple Feature Disentanglement for One-Stage Adaptive Object Detection (AAAI 2024 · similarity 0.92)
- Single-Domain Generalized Object Detection in Urban Scene via Cyclic-Disentangled Self-Distillation (CVPR 2022 · similarity 0.91)
- Domain-Specific Suppression for Adaptive Object Detection (CVPR 2021 · similarity 0.90)
- Non-exemplar Domain Incremental Object Detection via Learning Domain Bias (AAAI 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.
