Research map / Vision-language and multimodal / Human and embodied modeling
Freeview Sketching: View-Aware Fine-Grained Sketch-Based Image Retrieval
Aneeshan Sain, Pinaki Nath Chowdhury, Subhadeep Koley, Ayan Kumar Bhunia, 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 disentangled view feature and view-specific retrieval objective in place of a single shared feature | FG-SBIR baseline · pilot study, chairs, view-mismatched gallery | Acc@1: 58.25 then a drop of 27.15% | Hurt |
| add view-specific triplet loss L_Tri^VS (ablation) | Ours VGG-16 · Chairs view-specific | Top-1: 21.12 -> 60.71 | Improved |
| add view-agnostic triplet loss L_Tri^VA (ablation) | Ours VGG-16 · Chairs view-agnostic | mAP@all: 0.104 -> 0.615 | Improved |
| add instance-consistency loss on multi-view projections (ablation) | Ours VGG-16 · Chairs view-agnostic | mAP@all: 0.520 -> 0.615 | Improved |
| add view-consistency loss between sketch and photo views (ablation) | Ours VGG-16 · Chairs view-specific | Top-1: 55.69 -> 60.71 | Improved |
| add cross-view reconstruction loss L_VR (ablation) | Ours VGG-16 · Chairs view-specific | Top-1: 56.26 -> 60.71 | Improved |
| use multi-view 2D projections of 3D shapes during training (ablation) | Ours VGG-16 · Chairs view-specific | Top-1: 50.79 -> 60.71 | Improved |
| use one disentangled model instead of two separate models per paradigm (ablation) | Ours VGG-16 · Chairs view-agnostic | mAP@all: 0.421 -> 0.615 | Improved |
| use shared encoder instead of separate content and view encoders (ablation) | Ours VGG-16 · Chairs view-specific | Top-1: 54.32 -> 60.71 | Improved |
| reconstruct photo instead of edgemap in view reconstruction (ablation) | Ours VGG-16 · Chairs view-specific | Top-1: 57.68 -> 60.71 | Improved |
| replace PVT backbone with VGG-16 backbone (ablation) | Ours · Chairs view-agnostic | mAP@all: 0.689 -> 0.615 | Hurt |
| add colour augmentation on rendered projections (ablation) | Ours VGG-16 · Chairs | mAP / Acc@1: 0.065 mAP, 2.85% Acc@1 | Improved |
Closest papers by meaning
- Sketch Less for More: On-the-Fly Fine-Grained Sketch-Based Image Retrieval (CVPR 2020 · similarity 0.91)
- Sketch Me That Shoe (CVPR 2016 · similarity 0.90)
- Generalising Fine-Grained Sketch-Based Image Retrieval (CVPR 2019 · similarity 0.90)
- Sketch Down the FLOPs: Towards Efficient Networks for Human Sketch (CVPR 2025 · similarity 0.88)
- How to Handle Sketch-Abstraction in Sketch-Based Image Retrieval? (CVPR 2024 · similarity 0.88)
- You'll Never Walk Alone: A Sketch and Text Duet for Fine-Grained 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.
