Research map / Label-efficient and robust learning / Recognition and long-tail
Face Sketch Matching via Coupled Deep Transform Learning
Shruti Nagpal, Maneet Singh, Richa Singh, Mayank Vatsa, Afzel Noore, Angshul Majumdar
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 second DeepTransformer layer (L-CSSE) (ablation) | Semi-Coupled DeepTransformer · P1 (CSA) | Rank-10 accuracy (%): 34.1 -> 42.6 | Improved |
| add second DeepTransformer layer (L-CSSE) (ablation) | Semi-Coupled DeepTransformer · P2 (CSA) | Rank-10 accuracy (%): 37.7 -> 44.2 | Improved |
| add second DeepTransformer layer (L-CSSE) (ablation) | Semi-Coupled DeepTransformer · P3 (CSA) | Rank-10 accuracy (%): 28.9 -> 36.0 | Improved |
| add second DeepTransformer layer (L-CSSE) (ablation) | Symmetrically-Coupled DeepTransformer · P6 sketch-to-sketch | Rank-10 accuracy (%): 28.4 -> 31.5 | Improved |
| add second DeepTransformer layer (L-CSSE) (ablation) | Symmetrically-Coupled DeepTransformer · P7 sketch-to-sketch | Rank-10 accuracy (%): 31.5 -> 33.6 | Improved |
| add second DeepTransformer layer (Light CNN features) (ablation) | Semi-Coupled DeepTransformer · P1 (CSA) | Rank-10 accuracy (%): 30.0 -> 27.8 | Hurt |
| add DeepTransformer on top of L-CSSE features vs Euclidean distance (ablation) | Semi-Coupled DeepTransformer · P1 (CSA) | Rank-10 accuracy (%): 14.4 -> 42.6 | Improved |
| add DeepTransformer on top of L-CSSE features vs NNET alone (ablation) | Semi-Coupled DeepTransformer · P1 (CSA) | Rank-10 accuracy (%): 19.7 -> 42.6 | Improved |
| add DeepTransformer to existing features | Semi/Symmetrically-Coupled DeepTransformer · P1-P3, P6, P7 | Rank-10 accuracy (%): at least 10% as compared to Euclidean distance matching, and at most 22% | Improved |
| use L-CSSE class-specific features instead of other features | DeepTransformer · P1-P3, P6, P7 | Rank-10 accuracy (%): 42.6 vs 27.8 (DSIFT) | Improved |
| use greedy layer-by-layer training instead of joint training (ablation) | Semi-Coupled DeepTransformer · P1-P3 | Rank-10 accuracy / training time: 142 seconds, 12 seconds faster; same rank-10 accuracies | No clear effect |
Closest papers by meaning
- SketchTrans: Disentangled Prototype Learning With Transformer for Sketch-Photo Recognition (TPAMI 2023 · similarity 0.87)
- r-BTN: Cross-Domain Face Composite and Synthesis From Limited Facial Patches (AAAI 2018 · similarity 0.87)
- Face Sketch Synthesis From Coarse to Fine (AAAI 2018 · similarity 0.86)
- Learning to Sketch With Shortcut Cycle Consistency (CVPR 2018 · similarity 0.85)
- Cousin Network Guided Sketch Recognition via Latent Attribute Warehouse (AAAI 2019 · similarity 0.85)
- Coupled Deep Learning for Heterogeneous Face Recognition (AAAI 2018 · similarity 0.84)
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.
