Research map / Video / Video-language understanding
Multimodal Memorability: Modeling Effects of Semantics and Decay on Video Memorability
Anelise Newman, Camilo Fosco, Vincent Casser, Allen Lee, Barry McNamara, Aude Oliva
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 |
|---|---|---|---|
| combine frames, flow and video streams instead of video stream alone (ablation) | video stream I3D · Memento10k (test) | Spearman RC: 0.596 -> 0.659 | Improved |
| combine frames, flow and video streams instead of video stream alone (ablation) | video stream I3D · VideoMem (val) | Spearman RC: 0.492 -> 0.555 | Improved |
| add optical flow stream to frames stream (ablation) | frames stream · Memento10k (test) | Spearman RC: 0.595 -> 0.659 | Mixed |
| add caption generation to video stream (ablation) | video stream · Memento10k (test) | Spearman RC: 0.596 -> 0.602 | Improved |
| add caption generation to video stream (ablation) | video stream · VideoMem (val) | Spearman RC: 0.492 -> 0.512 | Improved |
| add triplet-loss semantic embedding to video stream (ablation) | video stream · Memento10k (test) | Spearman RC: 0.596 -> 0.599 | Improved |
| generate captions instead of learning a triplet-loss text embedding (ablation) | video stream · Memento10k (test) | Spearman RC: 0.599 -> 0.602 | Improved |
| use full SemanticMemNet instead of three-stream without captions (ablation) | three-stream model · Memento10k (test) | Spearman RC: 0.659 -> 0.663 | Improved |
| use full SemanticMemNet instead of three-stream without captions (ablation) | three-stream model · VideoMem (val) | Spearman RC: 0.555 -> 0.556 | No clear effect |
| use frames stream instead of video stream (ablation) | single stream · VideoMem (val) | Spearman RC: 0.492 -> 0.527 | Improved |
| use recurrent decay head instead of Mem-alpha, for rank correlation (ablation) | decay model · Memento10k (multi-lag) | Spearman RC: 0.604 -> 0.599 | Hurt |
| use recurrent decay head instead of Mem-alpha, for raw hit-rate prediction at t=80 (ablation) | decay model · Memento10k (lag t=80) | R^2: 0.227 -> 0.364 | Improved |
Closest papers by meaning
- A Review of Computational Memorability: A Benchmark Framework (IJCV 2026 · similarity 0.92)
- Episodic Memories Generation and Evaluation Benchmark for Large Language Models (ICLR 2025 · similarity 0.86)
- Before It Fades: Reinforcing Temporal Representations at Inference Time in VideoLLMs (NeurIPS 2026 · similarity 0.86)
- Bisecle: Binding and Separation in Continual Learning for Video Language Understanding (NeurIPS 2025 · similarity 0.86)
- Captured by Captions: On Memorization and its Mitigation in CLIP Models (ICLR 2025 · similarity 0.85)
- Quantifying Cross-Modality Memorization in Vision-Language Models (NeurIPS 2025 · similarity 0.85)
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.
