Research map / Label-efficient and robust learning / Recognition and long-tail
Learning Semantic-Specific Graph Representation for Multi-Label Image Recognition
Tianshui Chen, Muxin Xu, Xiaolu Hui, Hefeng Wu, Liang Lin
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 semantic decoupling and semantic interaction on top of ResNet-101 (ablation) | ResNet-101 · MS-COCO val | mAP: 80.3 -> 83.8 | Improved |
| add semantic decoupling (semantic-guided attention) (ablation) | SSGRL w/o SD · MS-COCO val | mAP: 80.9 -> 83.8 | Improved |
| add semantic interaction (graph propagation network) (ablation) | SSGRL w/o SI · MS-COCO val | mAP: 82.2 -> 83.8 | Improved |
| initialize graph nodes by plain average-pooled features instead of ResNet-101 head (w/o SD) (ablation) | ResNet-101 · MS-COCO val | mAP: 80.3 -> 80.9 | No clear effect |
| concatenate GloVe semantic vector to the image feature instead of semantic-guided attention (ablation) | SSGRL w/o SD · MS-COCO val | mAP: 80.9 -> 79.6 | Hurt |
| pre-train on MS-COCO before fine-tuning | SSGRL · PASCAL VOC 2007 | mAP: 93.4 -> 95.0 | Improved |
| pre-train on MS-COCO before fine-tuning | SSGRL · PASCAL VOC 2012 | mAP: 93.9 -> 94.8 | Improved |
| fuse two input scales (640 and 512) with ten-crop evaluation | SSGRL pre-trained · PASCAL VOC 2012 | mAP: 94.8 -> 95.4 | Improved |
| compare against ResNet-SRN under SSGRL | ResNet-SRN · MS-COCO val | mAP: 77.1 -> 83.8 | Improved |
| apply SSGRL to a 500-category setting | ResNet-SRN · Visual Genome 500 | mAP: 33.5 -> 36.6 | Improved |
| apply SSGRL to a hard category (toaster) (ablation) | ResNet-101 · MS-COCO val | AP: 24.7% improvement | Improved |
| apply SSGRL to an easy category (giraffe, zebra) (ablation) | ResNet-101 · MS-COCO val | AP | No clear effect |
Closest papers by meaning
- Knowledge-Guided Multi-Label Few-Shot Learning for General Image Recognition (TPAMI 2020 · similarity 0.91)
- Scene-Aware Label Graph Learning for Multi-Label Image Classification (ICCV 2023 · similarity 0.91)
- Transformer-Based Dual Relation Graph for Multi-Label Image Recognition (ICCV 2021 · similarity 0.91)
- Improving Multi-label Learning with Missing Labels by Structured Semantic Correlations (ECCV 2016 · similarity 0.91)
- Semantic-Aware Representation Blending for Multi-Label Image Recognition with Partial Labels (AAAI 2022 · similarity 0.90)
- HGDL: Heterogeneous Graph Label Distribution Learning (NeurIPS 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.
