Research map / Graph learning / Graph self-supervised learning
Self-Supervised Learning of Visual Graph Matching
Chang Liu, Shaofeng Zhang, Xiaokang Yang, Junchi Yan
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 SCGM self-supervised pretraining before BBGM finetuning on all VOC labels | BBGM · Pascal VOC all keypoints | F1 score: 55.3% -> 57.8% | Improved |
| add SCGM pretraining before NGMv2 finetuning on all VOC labels | NGMv2 · Pascal VOC all keypoints | F1 score: 54.0% -> 55.9% | Improved |
| add SCGM pretraining before BBGM finetuning | BBGM · Pascal VOC intersection | accuracy: 79.0% -> 80.6% | Improved |
| add SCGM pretraining before NGMv2 finetuning | NGMv2 · Pascal VOC intersection | accuracy: 80.1% -> 80.2% | No clear effect |
| pretrain SCGM on larger SPair-71k instead of VOC | SCGM + BBGM · Pascal VOC finetuned with 100% labels | accuracy: 57.76% -> 59.52% | Improved |
| pretrain SCGM on SPair-71k instead of VOC, with only 20% finetuning labels | SCGM + BBGM · Pascal VOC finetuned with 20% labels | accuracy: 51.72% -> 50.42% | Hurt |
| add contrastive loss (alpha 0.2 instead of 0.0) (ablation) | SCGM + BBGM · Pascal VOC all keypoints | F1 score: 34.0% -> 35.9% | Improved |
| raise mix rate lambda above 0.5 (0.4 to 0.6) (ablation) | SCGM + BBGM · Pascal VOC all keypoints | F1 score: 35.9% -> 25.8% | Hurt |
| add label mixing (lambda 0.4 instead of 0.0) (ablation) | SCGM + BBGM · Pascal VOC all keypoints | F1 score: 33.7% -> 35.9% | Improved |
| add graph augmentation on top of image augmentation (ablation) | SCGM + BBGM · Pascal VOC all keypoints | F1 score | Improved |
| add contrastive loss and label mixing to the augmentation stack (ablation) | SCGM + BBGM · Pascal VOC all keypoints | F1 score drop | Improved |
| add blur, gray-scale and color-jitter augmentations (ablation) | SCGM + BBGM · Pascal VOC all keypoints | F1 score drop | Improved |
Closest papers by meaning
- Contrastive Multi-View Representation Learning on Graphs (ICML 2020 · similarity 0.89)
- Learning deep graph matching with channel-independent embedding and Hungarian attention (ICLR 2020 · similarity 0.88)
- Deep Learning of Graph Matching (CVPR 2018 · similarity 0.88)
- A Simple Baseline for Weakly-Supervised Scene Graph Generation (ICCV 2021 · similarity 0.88)
- Learning Partial Graph Matching via Optimal Partial Transport (ICLR 2025 · similarity 0.86)
- Discrete Cycle-Consistency Based Unsupervised Deep Graph Matching (AAAI 2024 · 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.
