Research map / Architectures and efficiency / Attention and transformers
Glimpse-Attend-and-Explore: Self-Attention for Active Visual Exploration
Soroush Seifi, Abhishek Jha, Tinne Tuytelaars
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 contrastive stream (ablation) | Glimpse-Attend-and-Explore · SUN360 reconstruction | RMSE: 34.2 -> 33.8 | Improved |
| add contrastive stream (ablation) | Glimpse-Attend-and-Explore · MS-COCO reconstruction | RMSE: 43.7 -> 40.3 | Improved |
| add contrastive stream (ablation) | Glimpse-Attend-and-Explore · ADE20k segmentation | mPA (%): 52.3 -> 52.4 | No clear effect |
| add contrastive stream (ablation) | Glimpse-Attend-and-Explore · COCO-Stuff segmentation | mPA (%): 47.5 -> 47.8 | No clear effect |
| choose glimpses by self-attention heatmap instead of randomly (ablation) | Glimpse-Attend-and-Explore · SUN360 reconstruction | RMSE: 39.4 -> 33.8 | Improved |
| choose glimpses by self-attention heatmap instead of randomly (ablation) | Glimpse-Attend-and-Explore · MS-COCO reconstruction | RMSE: 47.8 -> 40.3 | Improved |
| choose glimpses by self-attention heatmap instead of randomly (ablation) | Glimpse-Attend-and-Explore · ADE20k segmentation | mPA (%): 51.5 -> 52.4 | Improved |
| replace self-attention with an extra conv channel under random glimpses (ablation) | Glimpse-Attend-and-Explore · ADE20k segmentation | mPA (%): 51.5 -> 52.4 | Improved |
| add attention-based glimpse selection to classifier (ablation) | Ours (Full) · SUN360 classification | accuracy (%): 49.6 -> 56.4 | Improved |
| add attention-based glimpse selection to no-decoder classifier (ablation) | Ours (No Decoder) · SUN360 classification | accuracy (%): 62.6 -> 67.2 | Improved |
| remove the reconstruction/segmentation decoder for classification (ablation) | Glimpse-Attend-and-Explore · SUN360 classification | accuracy (%): 56.4 -> 67.2 | Improved |
| use self-attention glimpse selection instead of Attend and Segment | Glimpse-Attend-and-Explore · ADE20k segmentation | mPA: 0.479 -> 0.524 | Improved |
Closest papers by meaning
- AdaGlimpse: Active Visual Exploration with Arbitrary Glimpse Position and Scale (ECCV 2024 · similarity 0.91)
- Learning to Look Around: Intelligently Exploring Unseen Environments for Unknown Tasks (CVPR 2018 · similarity 0.88)
- Mind the GAP: Glimpse-based Active Perception improves generalization and sample efficiency of visual reasoning (ICLR 2025 · similarity 0.87)
- CanViT: Toward Active-Vision Foundation Models (NeurIPS 2026 · similarity 0.87)
- Leveraging Separated World Model for Exploration in Visually Distracted Environments (NeurIPS 2024 · similarity 0.86)
- Focus-Then-Decide: Segmentation-Assisted Reinforcement Learning (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.
