AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

DarSwin: Distortion Aware Radial Swin Transformer

Akshaya Athwale, Arman Afrasiyabi, Justin Lagüe, Ichrak Shili, Ola Ahmad, Jean-François Lalonde

ICCV 2023 · Architectures and efficiency › Attention and transformers · approach: Token merging and masked pretraining · 10 citations (OpenAlex)

Read the paper

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 distortion-aware polar partition (lens information) (ablation)DarSwin-A · Distorted ImageNet-200, very low distortion testtop-1 accuracy: 52.2% -> 83.43%Improved
use distortion-aware sampling instead of uniform sampling (ablation)DarSwin-A · Distorted ImageNet-200, high distortion testtop-1 accuracy: 86.2% -> 88.27%Improved
add jittering of sample points (ablation)DarSwin-A · Distorted ImageNet-200, very low distortion testtop-1 accuracy: 81.5% -> 83.43%Improved
angular relative PE instead of Fourier polar PE (ablation)DarSwin-A · Distorted ImageNet-200, high distortion testtop-1 accuracy: 85.4% -> 88.27%Improved
increase radial sampling points from 2 to 10 (ablation)DarSwin-A · Distorted ImageNet-200, high distortion testtop-1 accuracy: 82.4% -> 88.27%Improved
merge/window along azimuth (DarSwin-A) instead of radius+azimuth (DarSwin-RA) (ablation)DarSwin · Distorted ImageNet-200, very low distortion testtop-1 accuracy: 79.7% -> 83.43%Improved
use DarSwin instead of Swin (trained on high distortion)Swin · Distorted ImageNet-200, test xi=0.4top-1 accuracy: 40.1 -> 91.39Improved
DarSwin vs Swin (undis) under projection model shift to polynomialDarSwin · Polynomial projection test settop-1 accuracy: 82.7% vs 85.9% (very low); 65.5% vs 60.3% (high)Mixed

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