Research map / Video / Video-language understanding
Learning Procedural-Aware Video Representations Through State-Grounded Hierarchy Unfolding
Jinghan Zhao, Yifei Huang, Feng Lu
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 State stage after Step (Path-3 vs Path-2) (ablation) | progressive pre-training · COIN, MLP head | step recognition acc: 43.06 -> 41.28 | Hurt |
| add return to Step after State (Path-5 vs Path-3) (ablation) | progressive pre-training · COIN, MLP head | step forecasting acc: 34.35 -> 38.07 | Improved |
| add State layer with step-return (Path-5 vs Path-2) (ablation) | progressive pre-training · COIN, MLP head | step recognition acc: 43.06 -> 44.54 | Improved |
| go from State directly back to Task (Path-4 vs Path-3) (ablation) | progressive pre-training · COIN, MLP head | task recognition acc: 80.73 -> 77.74 | Hurt |
| add final Task stage after Path-5 (Path-6) (ablation) | progressive pre-training · COIN, MLP head | step forecasting acc: 38.07 -> 36.94 | No clear effect |
| replace joint training with progressive pathway (ablation) | TSS pre-training · COIN, MLP head | step forecasting acc: 29.79 -> 38.07 | Improved |
| progressive Path-5 vs fusing three level features by average pooling (ablation) | TSS pre-training · COIN, MLP head | step forecasting acc: 36.88 -> 38.07 | Improved |
| use Path-5 instead of Paprika | pretraining method · COIN, Transformer head | step forecasting acc: 38.93 -> 40.40 | Improved |
| use Path-5 instead of Paprika | pretraining method · CrossTask, MLP head | task recognition acc: 89.65 -> 89.44 | Hurt |
| add Task step pretraining (ablation) | adapter pretraining · COIN, MLP head | task recognition acc: 2.09 -> 82.45 | Improved |
Closest papers by meaning
- VEDIT: Latent Prediction Architecture For Procedural Video Representation Learning (ICLR 2025 · similarity 0.89)
- Grounding Video Models to Actions through Goal Conditioned Exploration (ICLR 2025 · similarity 0.88)
- Separating Knowledge and Perception with Procedural Data (ICML 2025 · similarity 0.87)
- Look, Remember and Reason: Grounded Reasoning in Videos with Language Models (ICLR 2024 · similarity 0.86)
- SCHEMA: State CHangEs MAtter for Procedure Planning in Instructional Videos (ICLR 2024 · similarity 0.86)
- Pre-trained Text-to-Image Diffusion Models Are Versatile Representation Learners for Control (NeurIPS 2024 · 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.
