Research map / Graph learning / Knowledge and temporal graphs
Affinity Graph Supervision for Visual Recognition
Check 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 |
|---|---|---|---|
| Add affinity mass supervision to the attention network | RelNet on VOC07 · VOC07 | mAP@all (%): 47.7 ± 0.1 -> 48.2 ± 0.1 | Helped |
| Add affinity mass supervision to the attention network | RelNet on mini COCO · mini COCO | avg mAP (%): 27.5 -> 27.9 | Helped |
| Add affinity supervision to the scene attention network | MIT67 scene categorization model · MIT67 | accuracy (%): 77.1 ± 0.2 -> 80.2 ± 0.3 | Helped |
| Add mini-batch affinity supervision | ResNet models · CIFAR-10, CIFAR-100, and Tiny ImageNet | accuracy (%) | Helped |
| Use focal affinity loss instead of smooth L1 or L2 (ablation) | RelNet on VOC07 · VOC07 | recall@5k (%): 60.3 ± 0.3 / 64.6 ± 0.5 -> 69.9 ± 0.3 | Helped |
| Set focal loss γ=5 instead of γ=2 (ablation) | RelNet on VOC07 · VOC07 | recall@5k (%): 69.9 ± 0.3 -> 66.8 ± 0.2 | Hurt |
| Set focal loss γ=5 instead of γ=2 (ablation) | RelNet on VOC07 · VOC07 | mAP@all (%): 48.2 ± 0.1 -> 48.6 ± 0.1 | Helped |
| Set focal loss γ=0 instead of γ=2 (ablation) | RelNet on VOC07 · VOC07 | recall@5k (%): 69.9 ± 0.3 -> 62.1 ± 0.3 | Hurt |
| Use a different-instance target that also selects same-category objects (ablation) | RelNet on VOC07 · VOC07 | avg mAP (%): 48.2 ± 0.2 -> 47.6 ± 0.1 | Hurt |
| Replace matrix-wise focal target mass with binary cross-entropy over individual edges (ablation) | Affinity supervision for relationship proposals · VOC07 relationship recovery | relationship recall | Hurt |
| Use matrix-wise target mass with a focal term rather than the alternative mass/loss formulations (ablation) | Affinity supervision for relationship proposals · VOC07 relationship recovery | relationship recall | Helped |
Closest papers by meaning
- LatentGNN: Learning Efficient Non-local Relations for Visual Recognition (ICML 2019 · similarity 0.88)
- Learning Affinity via Spatial Propagation Networks (NeurIPS 2017 · similarity 0.87)
- Learning deep graph matching with channel-independent embedding and Hungarian attention (ICLR 2020 · similarity 0.86)
- ResMatch: Residual Attention Learning for Feature Matching (AAAI 2024 · similarity 0.86)
- Graph-Context Attention Networks for Size-Varied Deep Graph Matching (CVPR 2022 · similarity 0.86)
- Neural Motifs: Scene Graph Parsing With Global Context (CVPR 2018 · 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.