Research map / Detection and segmentation / Open-set and few-shot detection
Sparse-Shot Learning With Exclusive Cross-Entropy for Extremely Many Localisations
Andreas Panteli, Jonas Teuwen, Hugo Horlings, Efstratios Gavves
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 cross-entropy by exclusive cross-entropy | YOLLO · TNBC detection (30% annotations) | F1: 0.65 -> 0.70 | Improved |
| replace cross-entropy by exclusive cross-entropy | YOLLO · TNBC detection (60% annotations) | F1: 0.7 -> 0.75 | Improved |
| add focal reweighing on unannotated terms of exclusive cross-entropy (ablation) | Exclusive cross-entropy · TNBC detection (30% annotations) | F1: 0.70 -> 0.74 | Improved |
| add focal reweighing on unannotated terms of exclusive cross-entropy (ablation) | Exclusive cross-entropy · TNBC detection (60% annotations) | F1: 0.75 -> 0.80 | Improved |
| add focal loss to standard cross-entropy (ablation) | Cross-entropy · TNBC detection (30% annotations) | F1: 0.65 -> 0.36 | Hurt |
| add focal loss to standard cross-entropy (ablation) | Cross-entropy · TNBC detection (60% annotations) | F1: 0.7 -> 0.49 | Hurt |
| add weak supervision pseudo-labels (tau 0.75) (ablation) | Cross-entropy · TNBC detection (30% annotations) | F1: 0.65 -> 0.64 | No clear effect |
| add weak supervision pseudo-labels (tau 0.50) (ablation) | Cross-entropy · TNBC detection (60% annotations) | F1: 0.7 -> 0.64 | Hurt |
| use sigmoid annealing of rho instead of fixed rho 0.75 (ablation) | Exclusive cross-entropy + focal loss · TNBC detection (30% annotations) | F1: 0.68 -> 0.74 | Improved |
| use sigmoid annealing instead of linear annealing (ablation) | Exclusive cross-entropy + focal loss · TNBC detection (60% annotations) | F1: 0.73 -> 0.80 | Improved |
| use exclusive cross-entropy instead of cross-entropy | Unet · Segmentation on nine datasets with sparse annotations | DICE: up to 85% | Improved |
| use exclusive cross-entropy instead of cross-entropy | YOLLO · TIL localisation | exclusive recall: 0.85 -> 0.88 | Improved |
Closest papers by meaning
- FOCUS: Knowledge-enhanced Adaptive Visual Compression for Few-shot Whole Slide Image Classification (CVPR 2025 · similarity 0.84)
- beta-FFT: Nonlinear Interpolation and Differentiated Training Strategies for Semi-Supervised Medical Image Segmentation (CVPR 2025 · similarity 0.83)
- Learned Representation-Guided Diffusion Models for Large-Image Generation (CVPR 2024 · similarity 0.83)
- Are nuclear masks all you need for improved out-of-domain generalisation? A closer look at cancer classification in histopathology (NeurIPS 2024 · similarity 0.83)
- Boltzmann Attention Sampling for Image Analysis with Small Objects (CVPR 2025 · similarity 0.83)
- Weakly Supervised Gland Segmentation with Class Semantic Consistency and Purified Labels Filtration (AAAI 2025 · similarity 0.83)
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.
