Research map / Video / Action recognition
Long Movie Clip Classification with State-Space Video Models
Md Mohaiminul Islam, Gedas Bertasius
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 |
|---|---|---|---|
| replace self-attention with S4 layers in the long-range decoder | Long Sequence Transformer · LVU Relation | top-1 accuracy: 52.38 -> 57.14 | Improved |
| replace self-attention with S4 layers in the long-range decoder | Long Sequence Transformer · LVU Scene | top-1 accuracy: 62.79 -> 67.44 | Improved |
| replace self-attention with S4 layers in the long-range decoder | Long Sequence Transformer · LVU Director | top-1 accuracy: 56.07 -> 62.61 | Improved |
| replace self-attention with S4 layers in the long-range decoder | Long Sequence Transformer · LVU Like ratio | MSE: 0.31 -> 0.26 | Improved |
| use S4 instead of self-attention at 11,760 tokens | Long Sequence Transformer · LVU efficiency | GPU memory (GB): 41.38 -> 5.15 | Improved |
| use S4 instead of self-attention at 11,760 tokens | Long Sequence Transformer · LVU efficiency | training samples/s: 1.88 -> 4.95 | Improved |
| add multi-scale pooling and channel scaling to vanilla S4 decoder (ablation) | ViS4mer · LVU content tasks | top-1 accuracy: 49.53 -> 55.12 | Improved |
| add multi-scale pooling and channel scaling to vanilla S4 decoder (ablation) | ViS4mer · LVU user engagement | MSE: 2.30 -> 1.94 | Improved |
| add pooling alone, without channel scaling (ablation) | vanilla S4 · LVU content tasks | top-1 accuracy: 49.53 -> 48.96 | Hurt |
| add channel scaling alone, without pooling (ablation) | vanilla S4 · LVU content tasks | top-1 accuracy: 49.53 -> 52.25 | Improved |
| replace S4 with Performer (ablation) | ViS4mer · LVU Relation | top-1 accuracy: 57.14 -> 50.00 | Hurt |
| replace S4 with Orthoformer (ablation) | ViS4mer · LVU Genre | top-1 accuracy: 54.71 -> 55.79 | No clear effect |
Closest papers by meaning
- SEAL: Semantic Attention Learning for Long Video Representation (CVPR 2025 · similarity 0.87)
- Re-thinking Temporal Search for Long-Form Video Understanding (CVPR 2025 · similarity 0.86)
- Enhancing Temporal Understanding in Video-LLMs through Stacked Temporal Attention in Vision Encoders (NeurIPS 2025 · similarity 0.86)
- Streaming Long Video Understanding with Large Language Models (NeurIPS 2024 · similarity 0.86)
- DejaVid: Encoder-Agnostic Learned Temporal Matching for Video Classification (CVPR 2025 · similarity 0.86)
- SALOVA: Segment-Augmented Long Video Assistant for Targeted Retrieval and Routing in Long-Form Video Analysis (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.
