Research map / Detection and segmentation / CNN features
Attend and Segment: Attention Guided Active Semantic Segmentation
Soroush Seifi, 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 |
|---|---|---|---|
| use 3-scale retina glimpses instead of full-resolution glimpses at 18% pixel budget (ablation) | glimpse-only agent · CityScapes | mean pixel accuracy: 71.9% -> 78.1% | Improved |
| use 3-scale retina glimpses instead of 2-scale at same pixel budget (ablation) | glimpse-only agent · CityScapes | mean pixel accuracy: 77.2% -> 78.1% | Improved |
| increase number of glimpses from 1 to 10 (ablation) | glimpse-only agent · CityScapes | mean pixel accuracy: 63% -> 78.1% | Improved |
| increase number of glimpses from 1 to 10 (ablation) | glimpse-only agent · CamVid | mean pixel accuracy: 68.2% -> 80.9% | Improved |
| increase number of glimpses from 1 to 10 (ablation) | glimpse-only agent · Kitti | mean pixel accuracy: 64.3% -> 76.1% | Improved |
| keep glimpse coverage ratio constant when doubling input size to 256x512 with 96x96 glimpses (ablation) | glimpse-only agent · CityScapes | mean pixel accuracy | No clear effect |
| choose next glimpse by predicted uncertainty instead of random location (ablation) | glimpse-only agent · CityScapes | mean pixel accuracy | Improved |
| choose next glimpse by uncertainty instead of horizon-only or restricted movement (ablation) | glimpse-only agent · CityScapes | mean pixel accuracy | Improved |
| add low-resolution 32x32 view of the whole scene at the first step (hybrid agent) (ablation) | glimpse-only agent · CityScapes, CamVid, Kitti | mean pixel accuracy | Improved |
| increase number of glimpses for the hybrid agent (ablation) | hybrid agent · CityScapes, CamVid, Kitti | gap to glimpse-only | Mixed |
| downscale input from full to 1/16 for scale-only agent (ablation) | scale-only agent · Kitti | mean pixel accuracy: 81.7 -> 75.5 | Hurt |
| use full-image U-net instead of glimpse-only agent with 18% pixels | glimpse-only agent · CityScapes | mean IoU: 0.508 -> 0.590 | Improved |
Closest papers by meaning
- Glimpse-Attend-and-Explore: Self-Attention for Active Visual Exploration (ICCV 2021 · similarity 0.88)
- ViewAL: Active Learning With Viewpoint Entropy for Semantic Segmentation (CVPR 2020 · similarity 0.88)
- Diffusion-Driven Two-Stage Active Learning for Low-Budget Semantic Segmentation (NeurIPS 2025 · similarity 0.87)
- Embodied Visual Active Learning for Semantic Segmentation (AAAI 2021 · similarity 0.87)
- CanViT: Toward Active-Vision Foundation Models (NeurIPS 2026 · similarity 0.86)
- Towards Continual Universal Segmentation (CVPR 2025 · 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.
