AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

FunnyBirds: A Synthetic Vision Dataset for a Part-Based Analysis of Explainable AI Methods

Robin Hesse, Simone Schaub-Meyer, Stefan Roth

ICCV 2023 · Label-efficient and robust learning › Recognition and long-tail · approach: Post-hoc explanation analysis · 29 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
take absolute value of Integrated Gradients attribution (ablation)Integrated Gradients · FunnyBirds, VGG16mean explainability score: 0.85 -> 0.83Hurt
take absolute value of IG: completeness versus contrastivity (ablation)Integrated Gradients · FunnyBirds, VGG16completeness / contrastivity: completeness 0.95 -> 0.97; contrastivity 0.92 -> 0.84Mixed
remove bias term (X-DNN vs ResNet-50 with IG) (ablation)ResNet-50 · FunnyBirdscompleteness, correctness, contrastivity: completeness 0.86 -> 0.91; correctness 0.59 -> 0.6; contrastivity 0.98 -> 0.87Mixed
use B-cos network instead of standard ResNet-50 with IGResNet-50 · FunnyBirdsmean explainability score: 0.81 -> 0.82Improved
B-cos network effect on background independenceB-cos network · FunnyBirdsbackground independence: 1 -> 0.87Hurt
switch backbone from ResNet-50 to VGG16 for model-agnostic XAI methodsXAI methods · FunnyBirdsexplanation scores: IG 0.81 -> 0.85Improved
use BagNet architecture whose structure matches attribution mapsBagNet · FunnyBirdsmean explainability score: 0.9 vs 0.86 (next best)Improved
ProtoPNet prototypes as explanationProtoPNet · FunnyBirdsmean explainability score: 0.48Hurt
use coarser methods (Grad-CAM, RISE) instead of fine-grained gradient methodsGrad-CAM, RISE · FunnyBirds, ResNet-50completeness: IG 0.86 vs Grad-CAM 0.74 and RISE 0.7Hurt
use LIME as explanation methodLIME · FunnyBirdscorrectness and contrastivity: 0 and 0Hurt

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