Research map / Detection and segmentation / Open-set and few-shot detection
Attentive Semantic Alignment with Offset-Aware Correlation Kernels
Paul Hongsuck Seo, Jongmin Lee, Deunsol Jung, Bohyung Han, Minsu Cho
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 proposed attentive alignment instead of GeoCNN | Proposed · Taniai (average) | flow accuracy: 0.67 -> 0.70 | Improved |
| use proposed model instead of GeoCNN on FG3DCar | Proposed · Taniai FG3DCar | flow accuracy: 0.85 -> 0.87 | Improved |
| use proposed model instead of GeoCNN on PASCAL subset | Proposed · Taniai PASCAL | flow accuracy: 0.53 -> 0.55 | Improved |
| use proposed model instead of GeoCNN on Caltech-101 IoU | Proposed · Caltech-101 | IoU: 0.56 -> 0.57 | Improved |
| use proposed model instead of GeoCNN on Caltech-101 LT-ACC | Proposed · Caltech-101 | LT-ACC: 0.79 -> 0.80 | Improved |
| use proposed model instead of GeoCNN on Caltech-101 localization error | Proposed · Caltech-101 | LOC-ERR: 0.28 -> 0.28 | No clear effect |
| combine TPS with affine instead of affine only (ImageNet objects) (ablation) | Proposed · Caltech-101 (ImageNet objects) | IoU: 0.54 -> 0.58 | Improved |
| combine affine with TPS instead of TPS only (ImageNet objects) (ablation) | Proposed · Caltech-101 (ImageNet objects) | IoU: 0.56 -> 0.58 | Improved |
| combine TPS with affine instead of affine only (other objects) (ablation) | Proposed · Caltech-101 (Other objects) | IoU: 0.52 -> 0.56 | Improved |
| test on objects unseen in ImageNet pretraining versus seen objects (ablation) | Proposed · Caltech-101 | IoU: 0.58 -> 0.56 | No clear effect |
| use correlation-based alignment on cross-class matching versus best prior model | Proposed · PASCAL Parts cross-class | wIoU: 37.5 -> 35.0 | No clear effect |
Closest papers by meaning
- Revisiting Efficient Semantic Segmentation: Learning Offsets for Better Spatial and Class Feature Alignment (ICCV 2025 · similarity 0.84)
- End-to-End Weakly-Supervised Semantic Alignment (CVPR 2018 · similarity 0.82)
- Text as Partial Constraint: Core–Residual Alignment for Robust Vision–Language Learning (NeurIPS 2026 · similarity 0.81)
- GOAL: Global-local Object Alignment Learning (CVPR 2025 · similarity 0.81)
- Seeing the Image: Prioritizing Visual Correlation by Contrastive Alignment (NeurIPS 2024 · similarity 0.80)
- AlignVLM: Bridging Vision and Language Latent Spaces for Multimodal Document Understanding (NeurIPS 2025 · similarity 0.80)
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.
