Research map / Video / Video segmentation and tracking
Introducing Decomposed Causality with Spatiotemporal Object-Centric Representation for Video Classification
Yachong Zhang, Lei Meng, Shuo Xu, Zhuang Qi, Wei Wu, Lei Wu, Xiangxu Meng
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 STEM module (ablation) | Uniformerv2-B base · MSR-VTT | Top-1 accuracy: 60.73 -> 62.23 | Improved |
| add STEM module (ablation) | Uniformerv2-B base · ActivityNet | Top-1 accuracy: 86.77 -> 88.58 | Improved |
| add compositional structure-aware attention (CSA) (ablation) | SurdCRL w/ Uniformerv2-B · MSR-VTT | Top-1 accuracy: 62.23 -> 62.50 | Improved |
| add adaptive sampling (AS) (ablation) | SurdCRL w/ Uniformerv2-B · ActivityNet | Top-1 accuracy: 88.94 -> 89.10 | Improved |
| add back-door intervention (BDI) (ablation) | SurdCRL w/ Uniformerv2-B · MSR-VTT | Top-1 accuracy: 62.80 -> 63.51 | Improved |
| add front-door intervention (FDI) after BDI (ablation) | SurdCRL w/ Uniformerv2-B · MSR-VTT | Top-1 accuracy: 63.51 -> 63.82 | Improved |
| add front-door intervention (FDI) after BDI (ablation) | SurdCRL w/ Uniformerv2-B · ActivityNet | Top-1 accuracy: 89.65 -> 89.95 | Improved |
| use saliency mask instead of zero mask for confounder dictionary (ablation) | SurdCRL w/ Uniformerv2-B · MSR-VTT | Top-1 accuracy: 63.16 -> 63.82 | Improved |
| use saliency mask instead of boxes mask (ablation) | SurdCRL w/ Uniformerv2-B · ActivityNet | Top-1 accuracy: 89.85 -> 89.95 | Improved |
| increase sampling count N beyond 10 in adaptive sampling (ablation) | SurdCRL · MSR-VTT and ActivityNet | Top-1 accuracy | Hurt |
| replace ORViT with SurdCRL on Uniformerv2-B | Uniformerv2-B · ActivityNet | Top-1 accuracy: 88.21 -> 89.95 | Improved |
| replace ECRL with SurdCRL on Uniformerv2-B | Uniformerv2-B · MSR-VTT | Top-5 accuracy: 86.33 -> 86.62 (Top-5), ActivityNet Top-5 97.52 -> 98.01 | Mixed |
Closest papers by meaning
- Spatial-temporal Causal Inference for Partial Image-to-video Adaptation (AAAI 2021 · similarity 0.87)
- Explicit Modeling of Causal Factors and Confounders for Image Classification (AAAI 2026 · similarity 0.86)
- Causal-Inspired Multitask Learning for Video-Based Human Pose Estimation (AAAI 2025 · similarity 0.86)
- A Large-Scale Study on Unsupervised Spatiotemporal Representation Learning (CVPR 2021 · similarity 0.86)
- Multimodal Event Causality Reasoning with Scene Graph Enhanced Interaction Network (AAAI 2024 · similarity 0.85)
- Object centric architectures enable efficient causal representation learning (ICLR 2024 · 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.
