EgoBabyVLM: Benchmarking Cross-Modal Learning from Naturalistic Egocentric Video Data

2026-05-18

AI Summary

Current vision-language models struggle to effectively learn language grounding from semantically weakly aligned and sparse naturalistic first-person videos, and lack dedicated evaluation frameworks. This work proposes Machine-DevBench, a benchmark that automatically constructs vocabulary and syntactic tests spanning logarithmic frequency ranges based on the model’s training lexicon, thereby eliminating train-evaluation mismatch. It further introduces a comprehensive evaluation suite encompassing both multimodal language grounding and unimodal tasks, trained and assessed on first-person video data from infants and adults. The study reveals that existing models heavily rely on strongly aligned web-collected data and exhibit substantial performance degradation in weakly aligned settings, highlighting a fundamental divergence from human language acquisition mechanisms. To advance research in infant-like naturalistic contexts, the authors launch the EgoBabyVLM Challenge.

Abstract

Children acquire language grounding with remarkable robustness from limited visuo-linguistic input in ways that surpass today's best large multimodal models. Recent research suggests current vision-language models (VLMs) trained on curated web data fail to generalize to the sparse, weakly-aligned egocentric streams produced by wearable devices, embodied agents, and infant head-cams -- and no fixed evaluation pipeline exists for measuring progress on this regime. We train VLMs on datasets with varying degrees of semantic alignment between visual and linguistic inputs, including naturalistic infant and adult egocentric videos, and evaluate them with a comprehensive suite spanning multimodal language grounding and unimodal vision and language tasks. At the core of this suite is Machine-DevBench, a corpus-grounded benchmark of lexical and grammatical competence, automatically generated from the model's training vocabulary across logarithmic frequency bins to eliminate the train/eval mismatch and low statistical power of prior developmental benchmarks. Our results show that current VLM paradigms hinge on the tight semantic alignment of curated data and fail to exploit the weakly-aligned signal that dominates naturalistic egocentric input -- the very regime in which humans thrive. To motivate progress, we introduce the EgoBabyVLM Challenge to drive the development of models capable of grounded language learning from the kind of naturalistic data that human infants experience.

Problem

• egocentric video
• vision-language models
• language grounding
• weakly-aligned data
• developmental benchmark

Innovation

• egocentric vision
• vision-language models
• cross-modal learning
• language grounding
• developmental benchmarking