Research map / Vision-language and multimodal / Human and embodied modeling
Sketch Less for More: On-the-Fly Fine-Grained Sketch-Based Image Retrieval
Ayan Kumar Bhunia, Yongxin Yang, Timothy M. Hospedales, Tao Xiang, Yi-Zhe Song
Check 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 |
|---|---|---|---|
| train the sketch branch with RL over complete sketch episodes and the proposed reward | FG-SBIR retrieval · Chair-V2 | m@A: 77.18 -> 85.44 | Improved |
| train the sketch branch with RL over complete sketch episodes and the proposed reward | FG-SBIR retrieval · Shoe-V2 | m@A: 80.12 -> 85.38 | Improved |
| add a global Kendall-Tau reward to the local reward (ablation) | Ours · Chair-V2 | m@A: +1.11% | Improved |
| add a global Kendall-Tau reward to the local reward (ablation) | Ours · Shoe-V2 | m@A: +1.31% | Improved |
| replace full rank-list Kendall-Tau consistency with a constraint on paired-photo rank only (ablation) | Ours · Chair-V2 | m@A: decreases by 0.78% | Hurt |
| replace full rank-list Kendall-Tau consistency with a constraint on paired-photo rank only (ablation) | Ours · Shoe-V2 | m@A: decreases by 0.86% | Hurt |
| use actor-only PPO with clipping rather than alternative RL methods (ablation) | Ours · Chair-V2 | m@A: 81.54 -> 85.44 | Improved |
| use actor-only PPO with clipping rather than alternative RL methods (ablation) | Ours · Shoe-V2 | m@A: 83.47 -> 85.38 | Improved |
| replace proposed reward with reciprocal-rank reward only (ablation) | Ours · Chair-V2 | m@A: 84.33 -> 85.44 | Hurt |
| replace proposed reward with reciprocal-rank reward only (ablation) | Ours · Shoe-V2 | m@A: 84.07 -> 85.38 | Hurt |
| predict covariance with a separate fully connected layer instead of a standalone trainable diagonal matrix (ablation) | Ours · Chair-V2 | m@A: deteriorates by 5.64% | Hurt |
| replace CNN sketch branch with an RNN (ablation) | Ours · Chair-V2 | A@5: 76.34 -> 19.62 | Hurt |
Closest papers by meaning
- Sketch Me That Shoe (CVPR 2016 · similarity 0.93)
- Generalising Fine-Grained Sketch-Based Image Retrieval (CVPR 2019 · similarity 0.93)
- ARNet: Self-Supervised FG-SBIR with Unified Sample Feature Alignment and Multi-Scale Token Recycling (AAAI 2025 · similarity 0.91)
- You'll Never Walk Alone: A Sketch and Text Duet for Fine-Grained Image Retrieval (CVPR 2024 · similarity 0.90)
- Sketch Down the FLOPs: Towards Efficient Networks for Human Sketch (CVPR 2025 · similarity 0.90)
- How to Handle Sketch-Abstraction in Sketch-Based Image Retrieval? (CVPR 2024 · similarity 0.89)
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.
