Research map / Detection and segmentation / CNN features
UPLiFT: Efficient Pixel-Dense Feature Upsampling with Local Attenders
Matthew Walmer, Saksham Suri, Anirud Aggarwal, Abhinav Shrivastava
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 |
|---|---|---|---|
| iterate LiFT to pixel-dense instead of single 2x step plus bilinear | LiFT · COCO-Stuff segmentation | mIoU: 58.28 -> 57.42 | Hurt |
| use UPLiFT (iterative, Local Attender) instead of LiFT | DINOv2-S/14 · Cityscapes segmentation | mIoU: 61.98 -> 65.38 | Improved |
| use UPLiFT instead of cross-attention AnyUp | DINOv2-S/14 · ADE20K segmentation | mIoU: 42.25 -> 42.97 | Improved |
| use UPLiFT instead of cross-attention AnyUp | DINOv2-S/14 · Feature upsampling 448x448 | inference time: 146.7 -> 79.4 ms | Improved |
| use UPLiFT local-only attention instead of AnyUp for depth | DINOv2-S/14 · COCO-Stuff depth | delta1: 61.32 vs 61.16 | Mixed |
| use local attention instead of quadratic cross-attention at ~1500 tokens | UPLiFT · 624x624 images | inference time: 2.5-5x faster | Improved |
| use UPLiFT instead of CFM for latent upscaling | SD1.5 · COCO-5k 512->1024 | FID: 28.81 -> 24.23 | Improved |
| use UPLiFT instead of CFM for latent upscaling | SD1.5 · FacesHQ super-resolution | FID: 1.36 -> 5.20 | Hurt |
| use UPLiFT instead of CFM for super-resolution | SD1.5 VAE · LHQ super-resolution | PSNR: 25.69 -> 26.70 | Improved |
| use larger UPLiFT model for VAE latents | UPLiFT · generative tasks | feature distribution | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 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
- LiFT — LIFT: Learning 4D LiDAR Image Fusion Transformer for 3D Object Detection (CVPR 2022) · 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
- Upsample Anything: A Simple and Hard to Beat Baseline for Feature Upsampling (CVPR 2026 · similarity 0.90)
- JAFAR: Jack up Any Feature at Any Resolution (NeurIPS 2025 · similarity 0.86)
- Minimal Interaction Seperated Tuning: A New Paradigm for Visual Adaptation (CVPR 2025 · similarity 0.84)
- FeatUp: A Model-Agnostic Framework for Features at Any Resolution (ICLR 2024 · similarity 0.84)
- Improving Depth Completion via Depth Feature Upsampling (CVPR 2024 · similarity 0.84)
- CARE Transformer: Mobile-Friendly Linear Visual Transformer via Decoupled Dual Interaction (CVPR 2025 · 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.
