Research map / Detection and segmentation / Weakly supervised detection and segmentation
Rethinking the Route Towards Weakly Supervised Object Localization
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| separate localization training from classification training instead of jointly optimizing both tasks (ablation) | PSOL · CUB-200 | Top-1 Loc: 74.24 -> 74.97 | Helped |
| separate localization training from classification training instead of jointly optimizing both tasks (ablation) | PSOL · ImageNet-1k | Top-1 Loc: 54.48 -> 55.31 | Helped |
| use DDT-generated class-agnostic pseudo boxes rather than classification-dependent WSOL localization (ablation) | PSOL · CUB-200 | GT-Known Loc: 59.30 -> 84.55 | Helped |
| use class-agnostic DDT-VGG16 localization rather than VGG16 CAM (ablation) | ImageNet-1k · Top-1 Loc | Top-1 Loc: 42.80 -> 47.31 | Helped |
| train a localization regressor on pseudo boxes instead of using DDT boxes directly | VGG16 localization · ImageNet-1k | Top-1 Loc: 47.31 -> 50.89 | Helped |
| use a separate localization model rather than joint classification-localization training (ablation) | PSOL-InceptionV3 · CUB-200 | Top-1 Loc: 60.32 -> 65.51 | Helped |
| use a separate localization model rather than joint classification-localization training (ablation) | PSOL-ResNet50 · CUB-200 | Top-1 Loc: 68.17 -> 70.68 | Helped |
| use a separate localization model rather than joint classification-localization training (ablation) | PSOL-VGG16 · ImageNet-1k | Top-1 Loc: 48.83 -> 50.89 | Helped |
| use a separate localization model rather than joint classification-localization training (ablation) | PSOL-VGG-GAP · ImageNet-1k | Top-1 Loc: 47.37 -> 48.36 | Helped |
| retain VGG16's original fully connected structure rather than replace it with GAP and one classifier layer (ablation) | PSOL separate training · CUB-200 | Top-1 Loc: 59.29 -> 66.30 | Helped |
| transfer the ImageNet-trained localization model to CUB-200 without fine-tuning (ablation) | PSOL-DenseNet161-Sep · CUB-200 | GT-Known Loc: 92.54 -> 92.07 | No clear effect |
| combine PSOL-DenseNet161-Sep localization with EfficientNet-B7 classification rather than DPN-131 classification | PSOL-DenseNet161-Sep · ImageNet-1k | Top-1 Loc: 56.59 -> 58.00 | Helped |
Closest papers by meaning
- Evaluating Weakly Supervised Object Localization Methods Right (CVPR 2020 · similarity 0.95)
- Shallow Feature Matters for Weakly Supervised Object Localization (CVPR 2021 · similarity 0.93)
- Bridging the Gap Between Classification and Localization for Weakly Supervised Object Localization (CVPR 2022 · similarity 0.92)
- Weakly Supervised Object Localization As Domain Adaption (CVPR 2022 · similarity 0.92)
- Learning to Localize Objects with Noisy Labeled Instances (AAAI 2019 · similarity 0.91)
- Unveiling the Potential of Structure Preserving for Weakly Supervised Object Localization (CVPR 2021 · similarity 0.90)
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.