Research map / Vision-language and multimodal / Visual question answering
A Read-Write Memory Network for Movie Story Understanding
Seil Na, Sangho Lee, Jisung Kim, Gunhee Kim
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 read and write convolutional networks (ablation) | RWMN · MovieQA video+subtitle val | accuracy: 34.20 -> 38.67 | Improved |
| add read network on top of write network (ablation) | RWMN-noR · MovieQA video+subtitle val | accuracy: 36.50 -> 38.67 | Improved |
| add write network (ablation) | RWMN-noRW · MovieQA video+subtitle val | accuracy: 34.20 -> 36.50 | Improved |
| add query-dependent memory embedding (ablation) | RWMN · MovieQA video+subtitle val | accuracy: 38.17 -> 38.67 | Improved |
| add video input (ablation) | RWMN · MovieQA video+subtitle val | accuracy: 37.20 -> 38.67 | Improved |
| bagging 30 bootstrapped models | RWMN · MovieQA video+subtitle val | accuracy: 38.67 -> 38.37 | No clear effect |
| ensemble of 20 random-init models | RWMN · MovieQA video+subtitle val | accuracy: 38.67 -> 38.30 | No clear effect |
| use RWMN instead of MEMN2N | RWMN · MovieQA DVS val | accuracy: 33.0 -> 40.0 | Improved |
| use RWMN instead of MEMN2N | RWMN · MovieQA Plot Synopses val | accuracy: 40.6 -> 37.0 | Hurt |
| increase write stride from 30 to 60 (ablation) | RWMN write network · MovieQA video+subtitle val | accuracy: 38.6 -> 33.6 | Hurt |
| use two write layers instead of one (ablation) | RWMN · MovieQA video+subtitle val | accuracy: 38.6 -> 37.2 | Hurt |
| use three write layers (40,3,3),(10,3,3),(10,3,3) (ablation) | RWMN · MovieQA video+subtitle val | accuracy: 38.6 -> 37.9 | Hurt |
Closest papers by meaning
- Movie Question Answering: Remembering the Textual Cues for Layered Visual Contents (AAAI 2018 · similarity 0.90)
- Progressive Attention Memory Network for Movie Story Question Answering (CVPR 2019 · similarity 0.89)
- ReWind: Understanding Long Videos with Instructed Learnable Memory (CVPR 2025 · similarity 0.85)
- MA-LMM: Memory-Augmented Large Multimodal Model for Long-Term Video Understanding (CVPR 2024 · similarity 0.84)
- VideoLucy: Deep Memory Backtracking for Long Video Understanding (NeurIPS 2025 · similarity 0.84)
- CALVIN: Improved Contextual Video Captioning via Instruction Tuning (NeurIPS 2024 · 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.
