AI Paper Atlas

Research map / Vision-language and multimodal / CLIP and zero-shot

NAF: Zero-Shot Feature Upsampling via Neighborhood Attention Filtering

Loïck Chambon, Paul Couairon, Éloi Zablocki, Alexandre Boulch, Nicolas Thome, Matthieu Cord

CVPR 2026 · Vision-language and multimodal › CLIP and zero-shot · approach: Attribute-based zero-shot learning · 3 citations (OpenAlex)

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 changedWhereMeasuredEffect
use NAF instead of VFM-specific JAFARupsampler · Pascal VOC12mIoU gain over Nearest: +5.12 -> +5.58Improved
add context (3x3) branch to the pixel encoder (ablation)NAF encoder · Kitti360mIoU: 59.86 -> 60.41Improved
use dual-branch blocks instead of Inception-style blocks (ablation)NAF encoder · Kitti360mIoU: 59.56 -> 60.41Improved
increase guidance dimension C from 256 to 1024 (ablation)NAF encoder · Kitti360mIoU: 60.41 -> 61.05Improved
increase conv blocks L from 2 to 5 (ablation)NAF encoder · Kitti360mIoU: 60.41 -> 61.13Improved
add 1x1 conv after pooling of keys (ablation)NAF attention keys · CityscapesmIoU: 60.41 -> 58.38Hurt
pool keys (AvgPool) instead of bilinear interpolation (ablation)NAF attention keys · CityscapesmIoU: 59.66 -> 60.41Improved
add RoPE positional encoding (ablation)NAF attention · CityscapesmIoU: 57.07 -> 60.41Improved
use RoPE instead of Gaussian spatial kernel (ablation)NAF attention · CityscapesmIoU: 58.19 -> 60.41Improved
use NAF instead of JAFAR for video propagationupsampler · DAVISJ&F gain: +3.54 vs +3.37Mixed
replace bilinear with NAF in ProxyCLIPopen-vocab segmentation · Pascal VOCmIoU: +1.04 mean gainImproved
use NAF for Gaussian denoising instead of Restormerimage restoration · ImageNet Gaussian denoising sigma=0.1PSNR: 32.51 vs 32.12Hurt

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.

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

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy