Research map / Label-efficient and robust learning / Recognition and long-tail
Automatic Check-Out via Prototype-Based Classifier Learning from Single-Product Exemplars
Hao Chen, Xiu-Shen Wei, Faen Zhang, Yang Shen, Hui Xu, Liang Xiao
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 |
|---|---|---|---|
| replace FC box classifier with prototype-based classifier from exemplars (ablation) | Faster RCNN + FPN · RPC averaged | cAcc: 82.87% -> 85.12% | Improved |
| add discriminative re-ranking (soft) to exemplar prototype classifier (ablation) | PSP · RPC averaged | cAcc: 85.12% -> 85.93% | Improved |
| add multi-label loss (ablation) | PSP (exemplars, soft re-ranking) · RPC averaged | cAcc: 85.93% -> 86.69% | Improved |
| use soft instead of hard re-ranking (with multi-label loss) (ablation) | PSP exemplar prototypes · RPC averaged | cAcc: 85.90% -> 86.69% | Improved |
| use soft instead of hard re-ranking (check-out prototypes, with multi-label loss) (ablation) | PSP check-out prototypes · RPC averaged | cAcc: 85.90% -> 85.65% | Hurt |
| generate classifiers from single-product exemplars instead of check-out instances (ablation) | PSP · RPC averaged | cAcc: 85.65% -> 86.69% | Improved |
| use exemplar prototypes instead of check-out prototypes (no re-ranking) (ablation) | PSP · RPC averaged | cAcc: 85.31% -> 85.12% | No clear effect |
| add hard re-ranking to check-out prototype classifier (ablation) | PSP · RPC averaged | cAcc: 85.31% -> 85.60% | Improved |
| use 1 exemplar per class instead of 8 for prototype (ablation) | PSP · RPC averaged | cAcc: 86.69% -> 85.21% | Hurt |
| vary exemplar count N_s among 4, 8, 80, all (ablation) | PSP · RPC averaged | cAcc: gap less than 0.8% | No clear effect |
| vary penalty factor C and slack epsilon (ablation) | PSP soft re-ranking · RPC averaged | cAcc: difference should not exceed 0.7% | No clear effect |
| replace Faster RCNN with Cascade RCNN backbone detector | PSP · RPC averaged | cAcc: 86.69% -> 92.05% | Improved |
Closest papers by meaning
- Learning from discriminative feature feedback (NeurIPS 2018 · similarity 0.78)
- From Individual to Whole: Reducing Intra-class Variance by Feature Aggregation (IJCV 2022 · similarity 0.78)
- Self-Regulated Feature Learning via Teacher-Free Feature Distillation (ECCV 2022 · similarity 0.77)
- DiAPR: Dimensionally-Allocated Prototype Refinement for Non-Exemplar Class Incremental Learning (AAAI 2026 · similarity 0.76)
- PROCAT: Product Catalogue Dataset for Implicit Clustering, Permutation Learning and Structure Prediction (NeurIPS 2021 · similarity 0.76)
- Effortless Active Labeling for Long-Term Test-Time Adaptation (CVPR 2025 · similarity 0.74)
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.
