AI Paper Atlas

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

CAESAR: An Embodied Simulator for Generating Multimodal Referring Expression Datasets

Md Mofijul Islam, Reza Manuel Mirzaiee, Alexi Gladstone, Haley N Green, Tariq Iqbal

NeurIPS 2022 · Vision-language and multimodal › Human and embodied modeling · approach: Referring expression comprehension · 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
add gaze cue to verbal and visual input (V+NH -> V+G) (ablation)CLIP · CAESAR-XLaccuracy: 64.78 -> 83.01Improved
add pointing gesture instead of gaze (V+NH -> V+P) (ablation)ViLT · CAESAR-XLaccuracy: 64.86 -> 84.43Improved
use both gaze and pointing instead of gaze only (ablation)CLIP · CAESAR-XLaccuracy: 83.01 -> 75.21Hurt
use both gaze and pointing instead of gaze only (ablation)Late Fusion · CAESAR-XLaccuracy: 81.51 -> 76.00Hurt
use both gaze and pointing instead of gaze only (ablation)Dual-Encoder · CAESAR-XLaccuracy: 72.26 -> 74.87Improved
verbal utterance only, no nonverbal cues (ablation)BERT · CAESAR-XLaccuracy: 50.00Hurt
top view instead of exo view (ablation)Late Fusion · CAESAR-XLaccuracy: 76.51 -> 88.37Improved
top view instead of exo view (ablation)Dual-Encoder · CAESAR-XLaccuracy: 68.84 -> 71.04Improved
ego view instead of exo view (ablation)ViLT · CAESAR-XLaccuracy: 62.47 -> 85.43Improved
all three views instead of best single view (ablation)Late Fusion · CAESAR-XLaccuracy: 88.37 -> 76.00Hurt
all three views instead of best single view (ablation)CLIP · CAESAR-XLaccuracy: 78.97 -> 75.21Hurt
SUM fusion instead of CONCAT (ablation)CLIP · CAESAR-XLaccuracy: 75.21 -> 82.85Improved

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