Research map / Detection and segmentation / Weakly supervised detection and segmentation
On Label Granularity and Object Localization
Elijah Cole, Kimberly Wilber, Grant Van Horn, Xuan Yang, Marco Fornoni, Pietro Perona, Serge Belongie, Andrew Howard, Oisin Mac Aodha
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 |
|---|---|---|---|
| train on coarser label granularity instead of species labels | CAM · iNatLoc500 | MaxBoxAccV2: +5.1 MaxBoxAccV2 | Improved |
| train on coarser label granularity instead of species labels | CutMix · iNatLoc500 | MaxBoxAccV2: +6.6 MaxBoxAccV2 | Improved |
| coarsen labels all the way to phylum | WSOL methods · iNatLoc500 | MaxBoxAccV2 | Hurt |
| train on coarser labels (manufacturer/family instead of variant) | CAM · FGVC-Aircraft | MaxBoxAccV2: +4.3 MaxBoxAccV2 | Improved |
| train on coarser labels | CAM · CUB | MaxBoxAccV2: +4.4 MaxBoxAccV2 | Improved |
| coarsen labels using ImageNet hierarchy | CAM · ImageNet | MaxBoxAccV2 | No clear effect |
| aggregate localization outputs of similar fine categories (CAM-Agg) instead of vanilla CAM | CAM · iNatLoc500 | MaxBoxAccV2 | Hurt |
| train at family level with 50 images per category | CAM · iNatLoc500 | MaxBoxAccV2: 9200 images matches full species-level CAM (~15x fewer images) | Improved |
| 30 hyperparameter sets instead of 5 (ablation) | CAM species-level · iNatLoc500 | MaxBoxAccV2: 60.2 -> 60.8 | Improved |
| train 10 epochs instead of 2 (ablation) | CAM · iNatLoc500 | MaxBoxAccV2 | Improved |
| train with FSL-Det few-shot detection on full-supervision boxes instead of WSOL | FSL-Det vs CAM · iNatLoc500 | MaxBoxAccV2: 60.2 -> 83.6 | Improved |
Closest papers by meaning
- Pairwise Similarity Knowledge Transfer for Weakly Supervised Object Localization (ECCV 2020 · similarity 0.91)
- Evaluating Weakly Supervised Object Localization Methods Right (CVPR 2020 · similarity 0.91)
- Evaluation for Weakly Supervised Object Localization: Protocol, Metrics, and Datasets (TPAMI 2022 · similarity 0.91)
- Rethinking the Route Towards Weakly Supervised Object Localization (CVPR 2020 · similarity 0.90)
- Weakly Supervised Object Localization Using Size Estimates (ECCV 2016 · similarity 0.89)
- Object-Aware Instance Labeling for Weakly Supervised Object Detection (ICCV 2019 · similarity 0.89)
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.
