Research map / Detection and segmentation / Weakly supervised detection and segmentation
Evaluation for Weakly Supervised Object Localization: Protocol, Metrics, and Datasets
Junsuk Choe, Seong Joon Oh, Sanghyuk Chun, Seungho Lee, Zeynep Akata, Hyunjung Shim
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 HaS erasing augmentation | CAM · Total mean (3 datasets) | mean localization: -1.4 | Hurt |
| add ACoL adversarial erasing head | CAM · Total mean (3 datasets) | mean localization: -3.7 | Hurt |
| add SPG self-produced guidance | CAM · Total mean (3 datasets) | mean localization: -0.9 | Hurt |
| add ADL attention-based dropout | CAM · Total mean (3 datasets) | mean localization: -1.8 | Hurt |
| add CutMix augmentation | CAM · Total mean (3 datasets) | mean localization: -0.5 | Hurt |
| replace CAM with GC-Net | CAM · ImageNet | MaxBoxAccV2: -9.9 | Hurt |
| use few-shot fully supervised training instead of WSOL hyperparameter search | CAM / best WSOL · Total mean (3 datasets) | mean localization: 64.5 -> 72.3 | Improved |
| use max normalization instead of min-max for score maps (ablation) | WSOL methods · Total mean (3 datasets) | mean localization: CutMix +0.2, ACoL -5.4 | No clear effect |
| use CAM score maps instead of input attribution (VB) | input attribution methods · OpenImages | mPxAP: 54.8 -> 60.5 | Improved |
| add SmoothGrad to vanilla backprop (ablation) | VB · OpenImages | PxAP: 54.0 to 49.6 | Hurt |
| add Integrated Gradients to vanilla backprop (ablation) | VB · OpenImages | PxAP: 54.0 to 57.1 | Improved |
| add Integrated Gradients to guided backprop (ablation) | GB · OpenImages | PxAP: 53.4 to 53.9 | No clear effect |
Closest papers by meaning
- Evaluating Weakly Supervised Object Localization Methods Right (CVPR 2020 · similarity 0.99)
- Weakly Supervised Object Localization As Domain Adaption (CVPR 2022 · similarity 0.95)
- Rethinking the Route Towards Weakly Supervised Object Localization (CVPR 2020 · similarity 0.94)
- Normalization Matters in Weakly Supervised Object Localization (ICCV 2021 · similarity 0.94)
- Rethinking Class Activation Mapping for Weakly Supervised Object Localization (ECCV 2020 · similarity 0.94)
- Weakly Supervised Object Localization via Transformer with Implicit Spatial Calibration (ECCV 2022 · similarity 0.94)
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.
