Research map / Detection and segmentation / CNN features
Upsample Anything: A Simple and Hard to Beat Baseline for Feature Upsampling
Minseok Seo, Mark Hamilton, Changick Kim
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 per-image anisotropic Gaussian TTO upsampling instead of bilinear | DINOv2-S linear probe · COCO | mIoU: 60.43 -> 61.41 | Improved |
| use per-image anisotropic Gaussian TTO upsampling instead of bilinear | DINOv2-S linear probe · ADE20k | mIoU: 41.48 -> 42.95 | Improved |
| use per-image anisotropic Gaussian TTO upsampling instead of bilinear | DINOv2-S linear probe · PASCAL-VOC | mIoU: 81.27 -> 82.22 | Improved |
| upsample predicted probability maps instead of features | Upsample Anything · PASCAL-VOC | mIoU: 82.22 -> 84.57 | Improved |
| upsample predicted probability maps instead of features | Upsample Anything · Cityscapes | mIoU: 57.92 -> 56.36 | Hurt |
| feature upsampling methods versus bilinear under strong backbone and long head training | all upsamplers · Cityscapes | mIoU: 57.90 -> 57.92 | No clear effect |
| use Upsample Anything instead of bilinear for depth | DINOv2 frozen, DPT-style head · NYUv2 depth estimation | RMSE: 0.545 -> 0.498 | Improved |
| use LoftUp upsampling instead of bilinear | DINOv2 DPT head · NYUv2 depth estimation | RMSE: 0.545 -> 0.796 | Hurt |
| use Upsample Anything instead of bilinear for depth map upsampling | 32x32 to 512x512 depth · Middlebury | RMSE: 0.231 -> 0.209 | Improved |
| use Upsample Anything instead of bilinear for depth map upsampling | 32x32 to 512x512 depth · NYUv2 depth map upsampling | RMSE: 0.167 -> 0.214 | Hurt |
| add per-pixel anisotropic Gaussian kernels with TTO (GSJBU) over fixed-kernel JBU (ablation) | JBU · PASCAL-VOC | mIoU: 81.65 -> 82.22 | Improved |
| add spatial-range constraint (GSJBU) over Guided Linear Upsample (ablation) | Guided Linear Upsample · NYUv2 depth estimation | RMSE: 0.598 -> 0.498 | Improved |
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.
- LoftUp — LoftUp: Learning a Coordinate-Based Feature Upsampler for Vision Foundation Models (ICCV 2025) · better on 4 of 4 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
- AnyUp: Universal Feature Upsampling (ICLR 2026 · similarity 0.91)
- Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data (CVPR 2024 · similarity 0.86)
- TinySAM: Pushing the Envelope for Efficient Segment Anything Model (AAAI 2025 · similarity 0.85)
- FeatUp: A Model-Agnostic Framework for Features at Any Resolution (ICLR 2024 · similarity 0.85)
- JAFAR: Jack up Any Feature at Any Resolution (NeurIPS 2025 · similarity 0.85)
- Uncertainty-aware Fine-tuning of Segmentation Foundation Models (NeurIPS 2024 · similarity 0.84)
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.
