Research map / Vision-language and multimodal / Human and embodied modeling
Defining Image Memorability Using the Visual Memory Schema
Erdem Akagündüz, Adrian G. Borş, Karla K. Evans
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 |
|---|---|---|---|
| pool features with VMS maps instead of whole image (ablation) | combined GIST+SIFT+HoG SVR · VISCHEMA memorability prediction | Spearman rank correlation: 0.24 -> 0.41 | Improved |
| pool features with VMS maps instead of whole image (ablation) | HoG SVR · VISCHEMA memorability prediction | Spearman rank correlation: 0.24 -> 0.34 | Improved |
| pool features with eye-fixation maps (ablation) | combined features SVR · VISCHEMA memorability prediction | Spearman rank correlation: 0.24 -> 0.29 | Improved |
| pool features with GBVS saliency maps (ablation) | combined features SVR · VISCHEMA memorability prediction | Spearman rank correlation: 0.24 -> 0.25 | No clear effect |
| pool GIST features with eye-fixation maps (ablation) | GIST SVR · VISCHEMA memorability prediction | Spearman rank correlation: 0.19 -> 0.13 | Hurt |
| pool GIST features with VMS maps (ablation) | GIST SVR · VISCHEMA memorability prediction | Spearman rank correlation: 0.19 -> 0.10 | Hurt |
| pool pixel histograms with VMS maps (ablation) | Pixels SVR · VISCHEMA memorability prediction | Spearman rank correlation: 0.044 -> 0.085 | Improved |
| use VMS pooling on combined features, bottom-20 images | combined features SVR · VISCHEMA memorability prediction | Bottom-20 average empirical memorability (%): 32.15 -> 24.87 | Improved |
| use VMS pooling on combined features, top-20 images | combined features SVR · VISCHEMA memorability prediction | Top-20 average empirical memorability (%): 56.10 -> 61.25 | Improved |
| use l1-norm instead of l2-norm loss (ablation) | VMS reconstruction CNN · VMS reconstruction (all VISCHEMA) | Pearson correlation of reconstructed vs human VMS | Improved |
| train longer than 20 epochs (ablation) | VMS reconstruction CNN · VMS reconstruction (all VISCHEMA) | Pearson correlation of reconstructed vs human VMS: 0.588 at epoch 20 | Hurt |
| transfer from deeper layers (conv-5^2) of VGG-VD-19L instead of shallow layers (ablation) | VMS reconstruction CNN · VMS reconstruction (all VISCHEMA) | Pearson correlation of reconstructed vs human VMS: 0.588 | Improved |
Closest papers by meaning
- VisMem: Latent Vision Memory Unlocks Potential of Vision-Language Models (CVPR 2026 · similarity 0.88)
- Déjà Vu Memorization in Vision–Language Models (NeurIPS 2024 · similarity 0.86)
- Towards flexible perception with visual memory (ICML 2025 · similarity 0.86)
- Quantifying Cross-Modality Memorization in Vision-Language Models (NeurIPS 2025 · similarity 0.85)
- Exploring Local Memorization in Diffusion Models via Bright Ending Attention (ICLR 2025 · similarity 0.84)
- Do VLMs Perceive or Recall? Probing Visual Perception vs. Memory with Classic Visual Illusions (CVPR 2026 · similarity 0.84)
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.
