Research map / Detection and segmentation / CNN features
FOVEA: Foveated Image Magnification for Autonomous Navigation
Chittesh Thavamani, Mengtian Li, Nicolas Cebron, Deva Ramanan
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 explicit backward mapping of boxes to original space (ablation) | FOVEA · Argoverse-HD val, after finetuning | AP: 2.5 -> 28.1 | Improved |
| add anti-crop reflect-padding regularization (ablation) | FOVEA · Argoverse-HD val, after finetuning | AP: 26.9 -> 28.1 | Improved |
| warp with KDE S_I saliency before finetuning | Faster R-CNN · Argoverse-HD val, before finetuning | AP: 21.5 -> 24.1 | Improved |
| warp with KDE S_I saliency after finetuning | Faster R-CNN · Argoverse-HD val, after finetuning | AP: 24.2 -> 28.0 | Improved |
| replace S_D dataset prior with temporal prior S_I (ablation) | KDE warping · Argoverse-HD val, after finetuning | AP: 26.7 -> 28.0 | Improved |
| combine S_I and S_D into S_C (ablation) | KDE warping · Argoverse-HD val, after finetuning | AP: 28.0 -> 27.2 | No clear effect |
| use separable instead of nonseparable learned warp (ablation) | learned saliency · Argoverse-HD val, after finetuning | AP: 25.9 -> 27.2 | Improved |
| learn amplitude and bandwidth of KDE (LKDE) (ablation) | KDE S_I · Argoverse-HD val, after finetuning | AP: 28.0 -> 28.1 | No clear effect |
| add finetuning with warped inputs | KDE S_I · Argoverse-HD val | AP: 24.1 -> 28.0 | Improved |
| use ground-truth previous-frame saliency instead of detections (ablation) | KDE S_I · Argoverse-HD val, after finetuning | AP: 28.0 -> 29.6 | Improved |
| increase inter-frame jitter from 10 to 200 pixels (ablation) | KDE S_I · Argoverse-HD val, after finetuning | AP: 29.4 -> 23.6 | Hurt |
| add KDE S_I warping to the better-implementation + pseudo GT pipeline | full-stack streaming · Argoverse-HD streaming | streaming AP: 21.2 -> 23.0 | Improved |
Closest papers by meaning
- FOVI: A biologically-inspired foveated interface for deep vision models (ICML 2026 · similarity 0.90)
- Seeing More with Less: Human-like Representations in Vision Models (CVPR 2025 · similarity 0.88)
- SeqRank: Sequential Ranking of Salient Objects (AAAI 2024 · similarity 0.85)
- TransNeXt: Robust Foveal Visual Perception for Vision Transformers (CVPR 2024 · similarity 0.85)
- CanViT: Toward Active-Vision Foundation Models (NeurIPS 2026 · similarity 0.85)
- Foveated Instance Segmentation (CVPR 2025 · similarity 0.85)
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.
