Research map / Video / Video generation
Generative Action Tell-Tales: Assessing Human Motion in Synthesized Videos
Xavier Thomas, Youngsun Lim, Ananya Srinivasan, Audrey Zheng, Deepti Ghadiyaram
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 supervised contrastive loss (ablation) | action manifold encoder · TAG-Bench-v0 Action Consistency | Spearman correlation: 0.26 -> 0.61 | Improved |
| add supervised contrastive loss (ablation) | action manifold encoder · TAG-Bench-v0 Temporal Coherence | Spearman correlation: 0.38 -> 0.64 | Improved |
| add hard-negative temporal distortion loss (ablation) | action manifold encoder · TAG-Bench-v0 Action Consistency | Spearman correlation: 0.54 -> 0.61 | Improved |
| add hard-negative temporal distortion loss (ablation) | action manifold encoder · TAG-Bench-v0 Temporal Coherence | Spearman correlation: 0.57 -> 0.64 | Improved |
| add motion (temporal derivative) features (ablation) | action manifold encoder · TAG-Bench-v0 Action Consistency | Spearman correlation: 0.46 -> 0.61 | Improved |
| add 3D pose features (ablation) | action manifold encoder · TAG-Bench-v0 Action Consistency | Spearman correlation: 0.56 -> 0.61 | Improved |
| add 2D keypoint features (ablation) | action manifold encoder · TAG-Bench-v0 Action Consistency | Spearman correlation: 0.61 -> 0.61 | No clear effect |
| use ViT (TokenHMR) visual features instead of CLIP (ablation) | action manifold encoder · TAG-Bench-v0 Temporal Coherence | Spearman correlation: 0.39 -> 0.64 | Improved |
| add human-centric 3D/2D features to CLIP-only input (ablation) | action manifold encoder · TAG-Bench-v0 Action Consistency | Spearman correlation: 0.12 -> 0.51 | Improved |
| increase temporal window from 4 to 32 frames (ablation) | action manifold encoder · TAG-Bench-v0 Action Consistency | Spearman correlation: 0.39 -> 0.61 | Improved |
| increase temporal window beyond 32 frames (ablation) | action manifold encoder · TAG-Bench-v0 Action Consistency | Spearman correlation: 0.61 -> 0.56 | No clear effect |
| use attention-weighted per-frame fusion instead of no fusion (ablation) | action manifold encoder · TAG-Bench-v0 Action Consistency | Spearman correlation: 0.53 -> 0.61 | Improved |
Closest papers by meaning
- Direct Motion Models for Assessing Generated Videos (ICML 2025 · similarity 0.90)
- Aligning Human Motion Generation with Human Perceptions (ICLR 2025 · similarity 0.89)
- HumanVid: Demystifying Training Data for Camera-controllable Human Image Animation (NeurIPS 2024 · similarity 0.89)
- Video-Bench: Human-Aligned Video Generation Benchmark (CVPR 2025 · similarity 0.88)
- OpenHumanVid: A Large-Scale High-Quality Dataset for Enhancing Human-Centric Video Generation (CVPR 2025 · similarity 0.88)
- Auteur: Language-Driven Cinematographic Framing for Human-Centric Video Generation (NeurIPS 2026 · similarity 0.88)
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.
