AI Paper Atlas

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

SCLIP: Rethinking Self-Attention for Dense Vision-Language Inference

Feng Wang, Jieru Mei, Alan Yuille

ECCV 2024 · Vision-language and multimodal › CLIP and zero-shot · approach: Open-vocabulary alignment from VLMs · 141 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
replace CLIP self-attention with correlative self-attention (query-query and key-key similarity)CLIP ViT-B · 8 segmentation benchmarks avgmIoU (%): 14.1 -> 38.2Improved
use CSA with the pretrained Wq and Wk ensemble instead of identity projection (ablation)SCLIP · PASCAL VOCmIoU (%): 57.5 -> 59.1Improved
learn the projection matrix from 64 training samples (ablation)SCLIP · PASCAL VOCmIoU (%): 59.1 -> 60.4No clear effect
use random projection matrices (n=1 to 16) instead of pretrained (ablation)SCLIP · PASCAL VOCmIoU (%): 59.1 -> 57.1 (n=1), 58.1 (n=16)Hurt
sharpen CLIP attention via temperature instead of CSA (ablation)CLIP · VOC21 / Ctx59 / StuffmIoU (%): 18.8 -> 21.7 (tau=2) on VOC21; 13.3 -> 9.5 on Ctx59No clear effect
use local attention (window 3) instead of CSA (ablation)CLIP · VOC21mIoU (%): 59.1 -> 42.9Hurt
borrow early-layer attention maps instead of CSA (ablation)CLIP · VOC21mIoU (%): 59.1 -> 43.0 (layer 3)Hurt
use image size 336 / window 224 / stride 112 instead of prior 448/448/224 protocol (ablation)SCLIP · VOC21mIoU (%): 58.9 -> 59.1Improved
increase image size to 448 with window 224 (ablation)SCLIP · VOC21mIoU (%): 59.1 -> 60.4Improved
add PAMR post-processingSCLIP · 8 segmentation benchmarks avgmIoU (%): 38.2 -> 40.1Improved

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