Research map / Label-efficient and robust learning / Recognition and long-tail
Exploiting Unlabelled Photos for Stronger Fine-Grained SBIR
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 sketch and photo intra-modal triplet losses to cross-modal triplet training (ablation) | Ours-Strong · QMUL-Shoe-V2 | Acc.@1 (%): 43.28 -> 45.39 | Helped |
| Add EMA to the cross-modal and intra-modal triplet objectives (ablation) | Ours-Strong · QMUL-Shoe-V2 | Acc.@1 (%): 45.39 -> 46.50 | Helped |
| Add labelled-photo distillation to unlabelled-photo distillation (ablation) | Ours-Full · QMUL-Shoe-V2 | Acc.@1 (%): 46.50 -> 47.21 | Helped |
| Add sketch-to-photo contextual alignment distillation (ablation) | Ours-Full · QMUL-Shoe-V2 | Acc.@1 (%): 47.21 -> 48.35 | Helped |
| Add unlabelled-photo distillation to Ours-Strong | Ours-Full · QMUL-Shoe-V2 | Acc.@1 (%): 44.18 -> 48.35 | Helped |
| Add unlabelled-photo distillation to Ours-Strong | Ours-Full · QMUL-Chair-V2 | Acc.@1 (%): 71.22 -> 74.68 | Helped |
| Use a dedicated distillation token rather than the discriminative feature for distillation (ablation) | Ours-Full · QMUL-Shoe-V2 | Acc.@1 (%): 44.71 -> 48.35 | Helped |
| Process the distillation token at every PVT level via residual connections rather than only at the last stage (ablation) | Ours-Full · QMUL-Shoe-V2 | Acc.@1 (%): 45.31 -> 48.35 | Helped |
| Use structural photo augmentation for intra-modal triplet training instead of colour distortion (ablation) | Ours-Strong · QMUL-Shoe-V2 | Acc.@1 (%): 46.14 | Helped |
| Use structural photo augmentation for intra-modal triplet training instead of partial blurring (ablation) | Ours-Strong · QMUL-Shoe-V2 | Acc.@1 (%): 47.32 | Helped |
| Use structural photo augmentation for intra-modal triplet training instead of random sharpness shift (ablation) | Ours-Strong · QMUL-Shoe-V2 | Acc.@1 (%): 47.54 | Helped |
| Use photo intra-modal triplet loss rather than contrastive loss (ablation) | Ours-Strong · QMUL-Shoe-V2 | Acc.@1 (%): 45.68 | Helped |
Closest papers by meaning
- ARNet: Self-Supervised FG-SBIR with Unified Sample Feature Alignment and Multi-Scale Token Recycling (AAAI 2025 · similarity 0.92)
- Asymmetric Mutual Alignment for Unsupervised Zero-Shot Sketch-Based Image Retrieval (AAAI 2024 · similarity 0.91)
- Sketch Down the FLOPs: Towards Efficient Networks for Human Sketch (CVPR 2025 · similarity 0.90)
- You'll Never Walk Alone: A Sketch and Text Duet for Fine-Grained Image Retrieval (CVPR 2024 · similarity 0.90)
- Text-to-Image Diffusion Models are Great Sketch-Photo Matchmakers (CVPR 2024 · similarity 0.89)
- How to Handle Sketch-Abstraction in Sketch-Based Image Retrieval? (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.