Research map / Vision-language and multimodal / Multimodal LLMs
BabyVLM: Data-Efficient Pretraining of VLMs Inspired by Infant Learning
Shengao Wang, Arjun Chandra, Aoming Liu, Venkatesh Saligrama, Boqing Gong
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add child-directed transferred data to SAYCam (ablation) | BabyLLaVA-GPT2 · VTWT | accuracy: 0.6252 -> 0.6933 | Improved |
| add child-directed transferred data to SAYCam (ablation) | BabyLLaVA-GPT2 · Labeled-S | accuracy: 0.4195 -> 0.5364 | Improved |
| add child-directed transferred data to SAYCam (ablation) | CVCL · VTWT | accuracy: 0.6494 -> 0.7021 | Improved |
| add child-directed transferred data to SAYCam (ablation) | CVCL · Baby Winoground negative context | group score: 0.2246 -> 0.4493 | Improved |
| add child-directed transferred data to SAYCam (ablation) | CVCL · Baby Winoground positive context | group score: 0.5068 -> 0.4657 | Hurt |
| add child-directed transferred data to SAYCam (ablation) | CVCL · Labeled-S | accuracy: 0.6086 -> 0.5805 | Hurt |
| use transferred data instead of random general-domain data as the augmentation (ablation) | BabyLLaVA-GPT2 · VTWT | accuracy: 0.6553 -> 0.6933 | Improved |
| use transferred data instead of random general-domain data as the augmentation (ablation) | BabyLLaVA-GPT2 · SAYCam Caption | METEOR: 0.1778 -> 0.1592 | Hurt |
| use transferred data instead of random general-domain data as the augmentation (ablation) | CVCL · Baby Winoground overall | group score: 0.1068 -> 0.2027 | Improved |
| double the SAYCam data by relaxing the filter threshold (ablation) | BabyLLaVA-GPT2 · VTWT | accuracy: 0.6252 -> 0.6383 | Improved |
| add SAYCam anchor data to transferred data (ablation) | BabyLLaVA-GPT2 · VTWT | accuracy: 0.6239 -> 0.6933 | Improved |
| add image input to VTWT evaluation (ablation) | BabyLLaVA · VTWT | accuracy: 0.5360 -> 0.6252 | Improved |
Closest papers by meaning
- BabyVLM-V2: Toward Developmentally Grounded Pretraining and Benchmarking of Vision Foundation Models (CVPR 2026 · similarity 0.95)
- BabyVision: Visual Reasoning Beyond Language (ICML 2026 · similarity 0.90)
- DevBench: A multimodal developmental benchmark for language learning (NeurIPS 2024 · similarity 0.86)
- HoneyBee: Data Recipes for Vision-Language Reasoners (CVPR 2026 · similarity 0.86)
- Discovering Hidden Visual Concepts Beyond Linguistic Input in Infant Learning (CVPR 2025 · similarity 0.86)
- VLsI: Verbalized Layers-to-Interactions from Large to Small Vision Language Models (CVPR 2025 · similarity 0.86)
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.
