Research map / Label-efficient and robust learning / Recognition and long-tail
Multi-Label Image Recognition by Recurrently Discovering Attentional Regions
Zhouxia Wang, Tianshui Chen, Guanbin Li, Ruijia Xu, 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 |
|---|---|---|---|
| use learned attentional regions from spatial transformer instead of five object proposals (ablation) | recurrent memorized-attention model · PASCAL VOC 2007 test (single-crop 512) | mAP: 88.6 -> 90.4 | Improved |
| add anchor, scale and positive constraints on the transformer (ablation) | recurrent memorized-attention model · PASCAL VOC 2007 test (multi-crop 512) | mAP: 89.9 -> 91.3 | Improved |
| add anchor constraint to scale + positive (ablation) | attentional region constraints · PASCAL VOC 2007 test (multi-crop 512) | mAP: 90.2 -> 91.3 | Improved |
| add scale constraint to anchor + positive (ablation) | attentional region constraints · PASCAL VOC 2007 test (multi-crop 512) | mAP: 90.3 -> 91.3 | Improved |
| add positive constraint to anchor + scale (ablation) | attentional region constraints · PASCAL VOC 2007 test (multi-crop 512) | mAP: 90.4 -> 91.3 | Improved |
| add the three region constraints (ablation) | recurrent memorized-attention model · MS-COCO val (multi-crop 512, constraint ablation) | mAP: 71.5 -> 72.2 | Improved |
| add the three region constraints (ablation) | recurrent memorized-attention model · MS-COCO val (multi-crop 512, constraint ablation) | C-F1: 65.8 -> 66.5 | Improved |
| add the three region constraints (ablation) | recurrent memorized-attention model · MS-COCO val (multi-crop 512, constraint ablation) | O-F1: 70.9 -> 71.3 | Improved |
| use ten-crop instead of single-crop at scale 512 | recurrent memorized-attention model · PASCAL VOC 2007 test | mAP: 90.4 -> 91.3 | Improved |
| fuse two scales with ten-crop instead of one scale | recurrent memorized-attention model · PASCAL VOC 2007 test | mAP: 91.3 -> 91.9 | Improved |
| fuse two scales with ten-crop instead of one scale | recurrent memorized-attention model · MS-COCO val | mAP: 72.2 -> 73.4 | Improved |
| increase scale from 512 to 640 (single-crop) | recurrent memorized-attention model · MS-COCO val | mAP: 70.4 -> 70.5 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CNN-RNN — CNN-RNN: A Unified Framework for Multi-Label Image Classification (CVPR 2016) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Recurrent Attentional Reinforcement Learning for Multi-Label Image Recognition (AAAI 2018 · similarity 0.93)
- CNN-RNN: A Unified Framework for Multi-Label Image Classification (CVPR 2016 · similarity 0.92)
- Order-Free RNN With Visual Attention for Multi-Label Classification (AAAI 2018 · similarity 0.91)
- Multilabel Deep Visual-Semantic Embedding (TPAMI 2019 · similarity 0.91)
- Multi-Label Image Recognition With Graph Convolutional Networks (CVPR 2019 · similarity 0.90)
- Correlative and Discriminative Label Grouping for Multi-Label Visual Prompt Tuning (CVPR 2025 · 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.
