Research map / Detection and segmentation / Weakly supervised detection and segmentation
Weakly-Supervised Learning of Dense Functional Correspondences
Stefan Stojanov, Linan Zhao, Yunzhi Zhang, Daniel L. K. Yamins, Jiajun Wu
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 spatial contrastive loss to functional-only training (ablation) | Ours · Real evaluation dataset | PCK@23p: 0.336 -> 0.516 | Improved |
| add functional contrastive loss to spatial-only training (ablation) | Ours · Real evaluation dataset | AP@23p: 0.382 -> 0.691 | Improved |
| add mask loss (ablation) | Ours · Real evaluation dataset | Best F1@23p: 0.775 -> 0.808 | Improved |
| add mask loss (ablation) | Ours · Real evaluation dataset | PCK@23p: 0.516 -> 0.501 | Hurt |
| add CogVLM grounding to DINO | DINO · Real evaluation dataset | AP@23p: 0.382 -> 0.561 | Improved |
| prompt ManipVQA with function instead of part name (ablation) | ManipVQA + DINO · Real evaluation dataset | PCK@23p: 0.398 -> 0.309 | Hurt |
| distill into learned features instead of off-the-shelf DINO | DINO · Real evaluation dataset | Normalized Dist: 0.206 -> 0.153 | Improved |
| use LoRA finetuning and FiLM text conditioning (ablation) | Ours · Synthetic/Real evaluation | Normalized Dist: 0.172 to 0.181 | No clear effect |
| increase backbone from ViT-S to ViT-G (ablation) | Ours (with mask loss) · Real evaluation dataset | PCK@23p: 0.494 -> 0.523 | Improved |
| increase spatial loss weight from 1 to 10 (ablation) | Ours · Real evaluation dataset | PCK@23p: 0.443 -> 0.501 | Improved |
| increase mask loss weight from 1 to 10 (ablation) | Ours · Real evaluation dataset | AP@23p: 0.730 -> 0.719 | No clear effect |
| add half stride to ViT-B backbone (ablation) | Ours (with mask loss) · Real evaluation dataset | AP@10p: 0.360 -> 0.403 | Improved |
Closest papers by meaning
- Deep Functional Maps: Structured Prediction for Dense Shape Correspondence (ICCV 2017 · similarity 0.91)
- The Functional Correspondence Problem (ICCV 2021 · similarity 0.90)
- Zero-Shot Image Feature Consensus with Deep Functional Maps (ECCV 2024 · similarity 0.89)
- Continuous Surface Embeddings (NeurIPS 2020 · similarity 0.89)
- DenseMatcher: Learning 3D Semantic Correspondence for Category-Level Manipulation from a Single Demo (ICLR 2025 · similarity 0.88)
- Jamais Vu: Exposing the Generalization Gap in Supervised Semantic Correspondence (NeurIPS 2025 · similarity 0.86)
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.
