AI Paper Atlas

Research map / Vision-language and multimodal / Human and embodied modeling

Understanding and Evaluating Racial Biases in Image Captioning

Dora Zhao, Angelina Wang, Olga Russakovsky

ICCV 2021 · Vision-language and multimodal › Human and embodied modeling · approach: Neural image captioning · 97 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
add discriminability loss (DiscCap vs Att2in) (ablation)Att2in · COCO 2014 valCIDEr gap lighter-darker: 2.7 ± 0.7 -> 0.6 ± 0.7Mixed
use newer transformer-based models instead of FCFC · COCO 2014 valCIDEr gap lighter-darker: -0.8 ± 1.8 -> 3.2 ± 1.2Hurt
compare SPICE across skin tone groupsall six captioners · COCO 2014 valSPICE gap: 0.0 ± 0.1No clear effect
switch from human to generated captions (Transformer)Human captions · Controlled pairsembedding classifier AUC: 68.9 -> 54.2Improved
switch from FC to AoANet captionsFC · Controlled pairsvocabulary classifier AUC: 65.9 -> 68.0Hurt
switch from FC to Oscar (larger pretraining corpus)FC · Controlled pairsvocabulary classifier AUC: 65.9 -> 64.4Improved
switch from human to generated captionsHuman captions · Controlled pairsVADER sentiment gap: 1.5 -> 0.3Improved
add gender feature to distance-from-centerlogistic regression on layout · Skin tone classificationAUC: 56.6 -> 57.8Improved
match images by similarity and mask people (controlled set)visual classifier · Skin tone classificationAUC: 55.4 -> 44.7Improved
lighter versus darker skin images in human captionsHuman captions · COCO sentimentVADER compound: 0.073 -> 0.059Hurt
racial descriptors for individuals who appear Black versus whiteHuman captions · COCO train captionsdescriptor rate: 0.03% -> 0.54%Hurt

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