Research map / Label-efficient and robust learning / Recognition and long-tail
Neglected Free Lunch - Learning Image Classifiers Using Annotation Byproducts
Dongyoon Han, Junsuk Choe, Seonghyeok Chun, John Joon Young Chung, Minsuk Chang, Sangdoo Yun, Jean Y. Song, Seong Joon Oh
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 multitask regression of annotation-byproduct click location | ViT-B · ImageNet-1k val | top-1 accuracy: 81.6 -> 82.5 | Improved |
| add multitask regression of click location | ResNet18 · ImageNet-1k val | top-1 accuracy: 72.1 -> 72.2 | No clear effect |
| add click-location regression | ViT-B · ImageNet-A | accuracy: 26.1 -> 31.1 | Improved |
| add click-location regression | 7 models · ImageNet OOD (IN-A/C/O/R/Sketch) | accuracy: 30 of the 35 combinations | Mixed |
| add click-location regression | 7 models · SI-Score | accuracy: beating the baseline in all considered cases | Improved |
| add click-location regression | 7 models · BG Challenge | gap/accuracy: 12 out of 14 | Mixed |
| use real annotation byproduct instead of random point target (ablation) | ResNet50 · COCO | mAP: 73.6 -> 74.2 | Improved |
| use real annotation byproduct instead of random point target (ablation) | ViT-B · COCO | mAP: 74.5 -> 77.5 | Improved |
| add LUAB | ResNet50 · COCO | mAP: 73.0 -> 74.2 | Improved |
| add LUAB | ResNet50 · COCO | V^min: 47.6 -> 47.0 | Improved |
| add LUAB | ResNet50 · ImageNet WSOL | localization accuracy: 46.8 -> 48.4 | Improved |
| add LUAB | ResNet50 · COCO WSOL | mPxAP: 20.8 -> 21.5 | Improved |
Closest papers by meaning
- Towards Good Practices for Efficiently Annotating Large-Scale Image Classification Datasets (CVPR 2021 · similarity 0.88)
- Seeing Through the Human Reporting Bias: Visual Classifiers From Noisy Human-Centric Labels (CVPR 2016 · similarity 0.86)
- Visual Genome: Connecting Language and Vision Using Crowdsourced Dense Image Annotations (IJCV 2017 · similarity 0.86)
- L-WISE: Boosting Human Visual Category Learning Through Model-Based Image Selection and Enhancement (ICLR 2025 · similarity 0.85)
- Doomed to Re-Annotate, Forever: The ImageNet Story (NeurIPS 2026 · similarity 0.85)
- Modeling Collaborator: Enabling Subjective Vision Classification With Minimal Human Effort via LLM Tool-Use (CVPR 2024 · similarity 0.85)
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.
