Research map / Video / Video-language understanding
MECD+: Unlocking Event-Level Causal Graph Discovery for Video Reasoning
Tieyuan Chen, Huabin Liu, Yi Wang, Yihang Chen, Tianyao He, Chaofan Gan, Huanyu He, Weiyao Lin
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 caption loss L_C (ablation) | VGCM · MECD test set | accuracy: 64.8 -> 67.0 | Improved |
| add reconstruction loss L_V (ablation) | VGCM · MECD test set | accuracy: 65.1 -> 67.0 | Improved |
| add similarity loss L_S (ablation) | VGCM · MECD test set | accuracy: 65.3 -> 67.0 | Improved |
| add front-door adjustment (ablation) | VGCM · MECD test set | accuracy: 67.0 -> 68.7 | Improved |
| add counterfactual intervention (ablation) | VGCM · MECD test set | accuracy: 67.0 -> 69.3 | Improved |
| add context chain reasoning (ablation) | VGCM · MECD test set | accuracy: 71.2 -> 71.3 | No clear effect |
| add context chain reasoning (ablation) | VGCM · MECD test set | SHD: 4.19 -> 3.94 | Improved |
| use Videobert encoder instead of CLIP-L (ablation) | VGCM · MECD test set | accuracy: 69.99 -> 71.28 | Improved |
| add non-regressive complete graph reasoning (ablation) | VGCM · MECD inference speed | seconds/sample: 3.39 -> 0.76 | Improved |
| add front-door adjustment (ablation) | VGCM · MECD temporal illusory causality | accuracy drop at first relation: -3.3 -> -0.7 | Improved |
| add counterfactual intervention (ablation) | VGCM · MECD existence illusory test | similarity division: 1.04 -> 0.93 | Improved |
| use LLM reasoner instead of MLP head on the same encoder | VideoLLaVA vs LanguageBind · MECD test set | accuracy: 59.83 -> 67.12 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- Videobert — VideoBERT: A Joint Model for Video and Language Representation Learning (ICCV 2019) · better on 0 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- MECD: Unlocking Multi-Event Causal Discovery in Video Reasoning (NeurIPS 2024 · similarity 0.98)
- CausalChaos! Dataset for Comprehensive Causal Action Question Answering Over Longer Causal Chains Grounded in Dynamic Visual Scenes (NeurIPS 2024 · similarity 0.88)
- Two Causally Related Needles in a Video Haystack (NeurIPS 2025 · similarity 0.87)
- Multimodal Event Causality Reasoning with Scene Graph Enhanced Interaction Network (AAAI 2024 · similarity 0.85)
- Integrating Background Knowledge for Scalable Causal Discovery (NeurIPS 2026 · similarity 0.85)
- CausalTab: Pretraining Across Causal Environments for Tabular Causal Discovery (NeurIPS 2026 · 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.
