Research map / Vision-language and multimodal / Human and embodied modeling
Learning Dense Correspondences between Photos and Sketches
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 |
|---|---|---|---|
| photo-sketch pair contrastive pretraining of encoder (vs ImageNet-only) (ablation) | ResNet-18 encoder with STN warp estimator · PSC6k | PCK-5: 17.20 -> 56.38 | Helped |
| contrastive positives from same photo-sketch pair (vs same class pairs) (ablation) | encoder training · PSC6k | PCK-5: 54.73 -> 56.38 | Helped |
| contrastive learning on individual images (classical positives) instead of photo-sketch pairs (ablation) | encoder training · PSC6k | PCK-5: 56.38 -> 44.24 | Hurt |
| remove similarity loss L_sim (ablation) | warp estimator · PSC6k | PCK-5: 56.38 -> 17.46 without | Helped |
| direct similarity instead of perceptual similarity (ablation) | warp estimator · PSC6k | PCK-5: 49.31 -> 56.38 | Helped |
| add forward-backward consistency loss L_con (ablation) | warp estimator · PSC6k | PCK-5: 52.25 -> 56.38 | Helped |
| add weight function w (ablation) | warp estimator · PSC6k | PCK-5: 54.11 -> 56.38 | Helped |
| use multiple feature layers (vs single) (ablation) | warp estimator · PSC6k | PCK-5: 55.20 -> 56.38 | Helped |
| add conditional batch norm (ablation) | encoder · PSC6k | PCK-5: 55.84 -> 56.38 | Helped |
| ResNet-101 instead of ResNet-18 encoder | proposed method · PSC6k | PCK-5: 56.38 -> 58.00 | Helped |
| hold out 10 or 20 categories from training (ablation) | proposed method · PSC6k | PCK-5: 56.38 -> 56.05 (N=10), 55.88 (N=20) | No clear effect |
| proposed method vs WarpC-SemanticGLUNet (retrained) | photo-sketch correspondence · PSC6k | PCK-10: 79.70 -> 84.93 | Helped |
Closest papers by meaning
- Learning to Sketch With Shortcut Cycle Consistency (CVPR 2018 · similarity 0.90)
- Open Vocabulary Semantic Scene Sketch Understanding (CVPR 2024 · similarity 0.89)
- Sketch Down the FLOPs: Towards Efficient Networks for Human Sketch (CVPR 2025 · similarity 0.88)
- SketchFusion: Learning Universal Sketch Features through Fusing Foundation Models (CVPR 2025 · similarity 0.87)
- SketchINR: A First Look into Sketches as Implicit Neural Representations (CVPR 2024 · similarity 0.87)
- It's All About Your Sketch: Democratising Sketch Control in Diffusion Models (CVPR 2024 · similarity 0.87)
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.