Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Adaptive Fine-Grained Sketch-Based Image Retrieval
Ayan Kumar Bhunia, Aneeshan Sain, Parth Hiren Shah, Animesh Gupta, Pinaki Nath Chowdhury, Tao Xiang, Yi-Zhe Song
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 domain discriminator loss L_D (ablation) | adaptive FG-SBIR · Sketchy (unseen categories, k=5) | Acc@1: 26.3% -> 28.1% | Improved |
| add domain discriminator loss L_D (ablation) | adaptive FG-SBIR · QMUL-Shoe-V2 (unseen users, k=5) | Acc@1: 37.1% -> 38.3% | Improved |
| add semantic loss L_S (ablation) | adaptive FG-SBIR · Sketchy (unseen categories, k=5) | Acc@1: 23.7% -> 26.3% | Improved |
| add classification loss L_C (ablation) | adaptive FG-SBIR · Sketchy (unseen categories, k=5) | Acc@1: 16.5% -> 23.7% | Improved |
| add user-discrimination loss L_ud (ablation) | adaptive FG-SBIR · QMUL-Shoe-V2 (unseen users, k=5) | Acc@1: 35.8% -> 37.1% | Improved |
| meta-learn the inner-loop triplet margin instead of fixing it at 0.3 (ablation) | adaptive FG-SBIR · Sketchy (unseen categories, k=5) | Acc@1: 1.8% drop with fixed margin | Improved |
| meta-learn the inner-loop margin (ablation) | adaptive FG-SBIR · QMUL-Shoe-V2 (unseen users, k=5) | Acc@1: 0.06% boost | No clear effect |
| use a relational network to predict the margin instead of sum/average/max pooling (ablation) | adaptive FG-SBIR · Sketchy (unseen categories, k=5) | Acc@1: 1.2%/0.6%/0.8% drop with sum/average/max pooling | Improved |
| use fast-text instead of GloVe or word2vec for the semantic loss (ablation) | adaptive FG-SBIR · Sketchy (unseen categories, k=5) | Acc@1: 27.5% and 27.7% vs 28.1% | No clear effect |
| meta-learned adaptation instead of naive fine-tuning on the support set | baseline FG-SBIR · Sketchy (unseen categories, category-level) | Acc@1: 18.5% -> 28.1% | Improved |
| update only final embedding layer in inner loop (ANIL-style) instead of full MAML | MAML · Sketchy (unseen categories, category-level) | Acc@1: 22.8% -> 23.2% | Improved |
| add the regularizers alone without meta-learning | baseline FG-SBIR · Sketchy (unseen categories, category-level) | Acc@1: 18.4% -> 19.2% | Improved |
Closest papers by meaning
- Generalising Fine-Grained Sketch-Based Image Retrieval (CVPR 2019 · similarity 0.94)
- Sketch Less for More: On-the-Fly Fine-Grained Sketch-Based Image Retrieval (CVPR 2020 · similarity 0.92)
- StyleMeUp: Towards Style-Agnostic Sketch-Based Image Retrieval (CVPR 2021 · similarity 0.92)
- ARNet: Self-Supervised FG-SBIR with Unified Sample Feature Alignment and Multi-Scale Token Recycling (AAAI 2025 · similarity 0.91)
- How to Handle Sketch-Abstraction in Sketch-Based Image Retrieval? (CVPR 2024 · similarity 0.89)
- You'll Never Walk Alone: A Sketch and Text Duet for Fine-Grained 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.
