AI Paper Atlas

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

Ambiguous Images With Human Judgments for Robust Visual Event Classification

Kate Sanders, Reno Kriz, Anqi Liu, Benjamin Van Durme

NeurIPS 2022 · Vision-language and multimodal › Human and embodied modeling · approach: Human-like visual cognition in VLMs · 0 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
train on SQUID-E ambiguous images instead of standard high-certainty imagesResNet50 · SQUID-E 4-way event classificationavg accuracy: .61 -> .70Improved
train on SQUID-E instead of standard data, evaluated on standard dataResNet50 · Standard dataaccuracy: .93 -> .71Hurt
add geometric augmentation to standard-data trainingResNet50 · SQUID-E 4-way event classificationavg accuracy: .61 -> .64Improved
add noise injection and masking to standard-data trainingResNet50 · SQUID-E 4-way event classificationavg accuracy: .61 -> .64Improved
add photometric augmentation to standard-data trainingResNet50 · SQUID-E 4-way event classificationavg accuracy: .61 -> .57Hurt
add AugMix to standard-data trainingResNet50 · SQUID-E 4-way event classificationavg accuracy: .61 -> .60No clear effect
add combined augmentations to standard-data trainingResNet50 · SQUID-E 4-way event classificationstandard data accuracy: .93 -> .87Hurt
use focal loss instead of cross-entropy baselineResNet50 trained on SD · SQUID-E validationMSE vs human judgments: .15 -> .11Improved
use focal loss instead of cross-entropy baselineResNet50 trained on SD · SQUID-E validationECE: .58 -> .42Improved
use label smoothing instead of cross-entropy baselineResNet50 trained on SD · SQUID-E validationECE: .58 -> .46Improved
use Monte-Carlo Dropout instead of baselineResNet50 trained on SD · SQUID-E validationMSE vs human judgments: .15 -> .14No clear effect
use focal loss for model trained on SQUID-EResNet50 trained on SQUID-E · SQUID-E validationMSE vs human judgments: .16 -> .12Improved

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