Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Semantic-Aware Knowledge Preservation for Zero-Shot Sketch-Based Image Retrieval
Qing Liu, Lingxi Xie, Huiyu Wang, Alan L. Yuille
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 teacher-distillation loss L_T to benchmark loss (ablation) | SAKE (CSE-ResNet-50) · TU-Berlin Ext. ZS-SBIR 64-d | mAP@all: 0.353 -> 0.426 | Improved |
| add semantic-aware term (L_SAKE instead of L_T) (ablation) | SAKE (CSE-ResNet-50) · TU-Berlin Ext. ZS-SBIR 64-d | mAP@all: 0.426 -> 0.434 | Improved |
| add L_SAKE to benchmark loss (ablation) | ResNet-50 · TU-Berlin Ext. ZS-SBIR 64-d | mAP@all: 0.352 -> 0.413 | Improved |
| add L_SAKE to benchmark loss (ablation) | AlexNet · TU-Berlin Ext. ZS-SBIR 64-d | mAP@all: 0.267 -> 0.275 | Improved |
| add L_SAKE to benchmark loss (ablation) | CSE-ResNet-50 · TU-Berlin Ext. ZS-PBIR 64-d | mAP@all: 0.558 -> 0.683 | Improved |
| add L_T to benchmark loss (ablation) | ResNet-50 · TU-Berlin Ext. ZS-PBIR 64-d | mAP@all: 0.542 -> 0.666 | Improved |
| add semantic term lambda2 without teacher signal (lambda1=0) (ablation) | SAKE · TU-Berlin Ext. ZS-SBIR | mAP@all: 0.362 -> 0.370 | No clear effect |
| add semantic term lambda2=0.3 with teacher (lambda1=1) (ablation) | SAKE · TU-Berlin Ext. ZS-SBIR | mAP@all: 0.426 -> 0.434 | Improved |
| raise lambda2 from 0.3 to 3 (lambda1=1) (ablation) | SAKE · TU-Berlin Ext. ZS-SBIR | mAP@all: 0.434 -> 0.412 | Hurt |
| raise lambda_SAKE from 1 to 3 (ablation) | SAKE · TU-Berlin Ext. ZS-SBIR | mAP@all: 0.434 -> 0.429 | Hurt |
| raise lambda_SAKE from 0 to 3 on ImageNet categories (ablation) | SAKE · TU-Berlin Ext. ZS-PBIR (IN) | mAP@all: 0.545 -> 0.758 | Improved |
| fine-tune AlexNet with SAKE instead of benchmark loss only (ImageNet top-1 accuracy) (ablation) | AlexNet · ImageNet linear-probe top-1 (%) | mAP@all: 45.54 -> 51.39 | Improved |
Closest papers by meaning
- A Zero-Shot Framework for Sketch based Image Retrieval (ECCV 2018 · similarity 0.94)
- Doodle to Search: Practical Zero-Shot Sketch-Based Image Retrieval (CVPR 2019 · similarity 0.93)
- Transferable Coupled Network for Zero-Shot Sketch-Based Image Retrieval (TPAMI 2021 · similarity 0.92)
- Asymmetric Mutual Alignment for Unsupervised Zero-Shot Sketch-Based Image Retrieval (AAAI 2024 · similarity 0.90)
- ARNet: Self-Supervised FG-SBIR with Unified Sample Feature Alignment and Multi-Scale Token Recycling (AAAI 2025 · similarity 0.88)
- Text-to-Image Diffusion Models are Great Sketch-Photo Matchmakers (CVPR 2024 · similarity 0.88)
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.
