Research map / Video / Video-language understanding
Zero-Shot Text-to-Motion Evaluation using Video Language Models
Yuwen Ji, Donglin Wang, Yue Zhang
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 |
|---|---|---|---|
| select min-entropy view instead of max-entropy view (ablation) | VeMo InternVL3-14B · T2M meta-eval (MDM+MotionGPT) | AUC-ROC: 0.706 -> 0.720 | Improved |
| select min-entropy view instead of random view (ablation) | VeMo InternVL3-14B · T2M meta-eval (MDM+MotionGPT) | AUC-ROC: 0.711 -> 0.720 | Improved |
| min-entropy selection versus human-optimized view (ablation) | VeMo InternVL3-14B · T2M meta-eval (MDM+MotionGPT) | AUC-ROC: 0.723 vs 0.720 | No clear effect |
| increase K from 1 to 2 views (min-entropy) (ablation) | VeMo · T2M meta-eval (K views) | AUC-ROC: 0.706 -> 0.721 | Improved |
| increase K beyond 2 views (K=6) (ablation) | VeMo · T2M meta-eval (K views) | AUC-ROC: 0.721 -> 0.721 | No clear effect |
| increase K to 6 with max-entropy selection (ablation) | VeMo · T2M meta-eval (K views) | AUC-ROC: 0.706 -> 0.699 | Hurt |
| use 32 frames instead of 8 (ablation) | VeMo InternVL3-14B · T2M meta-eval (MDM+MotionGPT) | AUC-ROC: 0.720 -> 0.723 | Improved |
| use 128 frames instead of 32 (ablation) | VeMo InternVideo2.5 · T2M meta-eval (MDM+MotionGPT) | AUC-ROC: 0.688 -> 0.684 | Hurt |
| use InternVL3-14B instead of InternVL3-8B (ablation) | VeMo · T2M meta-eval (MDM+MotionGPT) | AUC-ROC: 0.687 -> 0.723 | Improved |
| use InternVL3-14B instead of InternVL3.5-14B (ablation) | VeMo · T2M meta-eval (MDM+MotionGPT) | AUC-ROC: 0.709 -> 0.723 | Improved |
| use joint video-text encoding VLM instead of ViCLIP independent encoding (ablation) | VeMo · T2M meta-eval (MDM+MotionGPT) | AUC-ROC: 0.559 -> 0.669 | Improved |
| paraphrase the input prompts (ablation) | VeMo InternVL3-14B · T2M meta-eval prompt robustness | AUC-ROC: 0.723 -> 0.708 | Hurt |
Closest papers by meaning
- Rethinking Human Evaluation Protocol for Text-to-Video Models: Enhancing Reliability, Reproducibility, and Practicality (NeurIPS 2024 · similarity 0.89)
- Evaluation of Text-to-Video Generation Models: A Dynamics Perspective (NeurIPS 2024 · similarity 0.89)
- T2V-CompBench: A Comprehensive Benchmark for Compositional Text-to-video Generation (CVPR 2025 · similarity 0.89)
- VimoRAG: Video-based Retrieval-augmented 3D Motion Generation for Motion Language Models (NeurIPS 2025 · similarity 0.88)
- Enhancing Motion in Text-to-Video Generation with Decomposed Encoding and Conditioning (NeurIPS 2024 · similarity 0.88)
- ViLMA: A Zero-Shot Benchmark for Linguistic and Temporal Grounding in Video-Language Models (ICLR 2024 · 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.
