Research map / Vision-language and multimodal / Human and embodied modeling
Are Vision Transformers More Data Hungry Than Newborn Visual Systems?
Lalit Pandey, Samantha Marie Waters Wood, Justin Newell Wood
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 |
|---|---|---|---|
| train ViT-CoT on more images (0, 1.25k, 10k, 80k) (ablation) | ViT-CoT · view-invariant object recognition | accuracy | Improved |
| increase ViT size from 3 to 6 or 9 heads/layers (ablation) | ViT-CoT · view-invariant object recognition | accuracy | No clear effect |
| reduce ViT to a single attention head/layer (ablation) | ViT-CoT · view-invariant object recognition | accuracy | Hurt |
| larger ViT-CoT versus smaller ViT-CoT data requirements (ablation) | ViT-CoT · training-set size sweep | accuracy | No clear effect |
| train linear classifier on only one viewpoint range (N_train = 1) (ablation) | ViT-CoT · view-invariant object recognition | accuracy | No clear effect |
| replace ViT-CoT with VideoMAE | ViT · view-invariant object recognition | accuracy | No clear effect |
| replace ViT-CoT with a CNN trained by SimCLR-CLTT | ViT-CoT · view-invariant object recognition | accuracy: 7.3\% | Improved |
| replace VideoMAE with a CNN trained by SimCLR-CLTT | VideoMAE · view-invariant object recognition | accuracy: 19.6\% | Improved |
| use 2-frame instead of 3-frame learning window (ablation) | ViT-CoT · view-invariant object recognition | accuracy | No clear effect |
| raise image resolution from 64x64 to 224x224 (ablation) | ViT-CoT and VideoMAE · view-invariant object recognition | accuracy | No clear effect |
| replace supervised linear classifier with unsupervised decoder (2AFC) (ablation) | ViT-CoT · 2AFC recognition | accuracy: 62.1\% vs 50\% chance | Improved |
Closest papers by meaning
- Learning to See Through a Baby's Eyes: Early Visual Diets Enable Robust Visual Intelligence in Humans and Machines (CVPR 2026 · similarity 0.88)
- Beyond the Doors of Perception: Vision Transformers Represent Relations Between Objects (NeurIPS 2024 · similarity 0.84)
- BabyVLM-V2: Toward Developmentally Grounded Pretraining and Benchmarking of Vision Foundation Models (CVPR 2026 · similarity 0.83)
- Learning from an Infant’s Visual Experience (AAAI 2024 · similarity 0.83)
- Prompt-CAM: Making Vision Transformers Interpretable for Fine-Grained Analysis (CVPR 2025 · similarity 0.83)
- Linearly Decomposing and Recomposing Vision Transformers for Diverse-Scale Models (NeurIPS 2024 · similarity 0.83)
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.
