AI Paper Atlas

Research map / 3D vision / 3D generation

Finding Differences Between Transformers and ConvNets Using Counterfactual Simulation Testing

Nataniel Ruiz, Sarah Adel Bargal, Cihang Xie, Kate Saenko, Stan Sclaroff

NeurIPS 2022 · 3D vision › 3D generation · approach: Monocular 3D object reconstruction · 2 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 ConvNext-T instead of Swin-TConvNext vs Swin · NVDtop-1 accuracy: 21.1 -> 24.7Improved
use ConvNext-S instead of Swin-S despite lower ImageNet accuracyConvNext vs Swin · NVDtop-1 accuracy: 22.1 -> 27.7Improved
use ConvNext-L-22k instead of Swin-L-22kConvNext vs Swin · NVDtop-5 accuracy: 61.4 -> 67.6Improved
use ConvNext-B instead of Swin-BConvNext vs Swin · NVDtop-1 accuracy: 22.7 -> 23.5Improved
pretrain on ImageNet-22k (Base size, ConvNext)ConvNext-B · NVDtop-1 accuracy: 23.5 -> 34.0Improved
pretrain on ImageNet-22k (Base size, Swin)Swin-B · NVDtop-1 accuracy: 22.7 -> 31.5Improved
increase network scale Tiny to SmallConvNext · NVDtop-1 accuracy: 24.7 -> 27.7Improved
increase network scale Small to Base (ImageNet-1k)ConvNext · NVDtop-1 accuracy: 27.7 -> 23.5Hurt
use Swin instead of ConvNext under random patch dropSwin vs ConvNext · ImageNet val patch dropaccuracyMixed
use Swin instead of ConvNext under maximal object occlusionSwin vs ConvNext · NVD occlusionproportion of correct conserved predictionsImproved
use ConvNext instead of Swin on small object scaleConvNext vs Swin · NVD scaleproportion of correct conserved predictionsImproved
use ConvNext instead of Swin on top viewsConvNext vs Swin · NVD top-to-frontal viewproportion of correct conserved predictions: ConvNext-L-22k around 73%Improved

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