AI Paper Atlas

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

Women also Snowboard: Overcoming Bias in Captioning Models

Lisa Anne Hendricks, Kaylee Burns, Kate Saenko, Trevor Darrell, Anna Rohrbach

ECCV 2018 · Vision-language and multimodal › Human and embodied modeling · approach: Neural image captioning · 415 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 Appearance Confusion Loss to Equalizer w/o ACL (ablation)Equalizer · MSCOCO-Biasgender error rate: 7.57 -> 7.02Improved
add Confident Loss to Equalizer w/o Conf (ablation)Equalizer · MSCOCO-Biasgender error rate: 9.62 -> 7.02Improved
add Appearance Confusion Loss to Equalizer w/o ACL (ablation)Equalizer · MSCOCO-Balancedgender error rate: 10.10 -> 8.10Improved
add Confident Loss to Equalizer w/o Conf (ablation)Equalizer · MSCOCO-Balancedgender error rate: 13.90 -> 8.10Improved
add Confident Loss to Equalizer w/o Conf (ablation)Equalizer · MSCOCO-Biasgender ratio difference: 0.09 -> -0.03Improved
add Appearance Confusion Loss to Equalizer w/o ACL (ablation)Equalizer · MSCOCO-Biasoutcome divergence between genders: 0.031 -> 0.018Improved
add Confident Loss to Equalizer w/o Conf (ablation)Equalizer · MSCOCO-Biasoutcome divergence between genders: 0.075 -> 0.018Improved
add Appearance Confusion Loss to Equalizer w/o ACL (ablation)Equalizer · MSCOCO-Balanced pointing game (Grad-CAM)pointing accuracy, all: 43.8 -> 47.5Improved
add Appearance Confusion Loss to Equalizer w/o ACL (ablation)Equalizer · masked images, man-to-woman ratiopredicted gender ratio (ideal 1.0): 3.45 -> 1.98Improved
rebalance training data by resampling women images (Balanced) vs fine-tuningBaseline-FT · MSCOCO-Biasgender error rate: 12.83 -> 12.85No clear effect
upweight gender words in cross entropy loss by 10Baseline-FT · MSCOCO-Bias vs MSCOCO-Balancedgender error rate: 12.83 -> 13.56 (Bias); 19.30 -> 16.30 (Balanced)Mixed
replace baseline with Equalizer (both losses) on shifted test distributionBaseline-FT · MSCOCO-Balancedgender error rate: 19.30 -> 8.10Improved

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