Research map / Vision-language and multimodal / CLIP and zero-shot
OmniFall: From Staged Through Synthetic to Wild, A Unified Multi-Domain Dataset for Robust Fall Detection
David Schneider, Zdravko Marinov, Moritz Mistol, Zeyun Zhong, Alexander Jaus, Rodi Düger, Rafael Baur, M. Saquib Sarfraz, Rainer Stiefelhagen
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 |
|---|---|---|---|
| Train on the unified OF-Staged instead of CMDFall alone (ablation) | VideoMAE-K400 · OF-ItW | Fall F1: 52.3 -> 61.2 | Improved |
| Train on the unified OF-Staged instead of CMDFall alone (ablation) | VideoMAE-K400 · OF-ItW | Fallen F1: 16.6 -> 39.4 | Improved |
| Train on synthetic OF-Syn instead of OF-Staged (ablation) | VideoMAE-K400 · OF-ItW | Fallen F1: 39.4 -> 42.3 | Improved |
| Add OF-Syn to OF-Staged training (ablation) | VideoMAE-K400 · OF-ItW | Fall F1: 64.2 -> 65.6 | Improved |
| Add OF-Staged to OF-Syn training (ablation) | VideoMAE-K400 · OF-ItW | Fallen F1: 42.3 -> 41.0 | No clear effect |
| Add in-the-wild training data (OF-ItW) to OF-Staged+OF-Syn (ablation) | VideoMAE-K400 · OF-ItW | Fall F1: 65.6 -> 76.0 | Improved |
| Train on synthetic OF-Syn only, test on staged CMDFall (ablation) | VideoMAE-K400 · CMDFall CS | BAcc: 83.3 -> 45.8 | Hurt |
| Test on in-the-wild instead of in-domain data | VideoMAE-K400 (Le2i-trained) · Fall F1 | F1: 100 -> 58.8 | Hurt |
| Use zero-shot 8B MLLM instead of small fine-tuned model for fall events | Qwen3-VL-8B / InternVL3.5-8B vs VideoMAE · OF-ItW | Fall F1: 67.8/68.0 vs 65.6 | No clear effect |
| Use fine-tuned VMAE trained on OF-Syn instead of zero-shot MLLMs for fallen state | VideoMAE-K400 · OF-ItW | Fallen F1: 29.2/31.0 -> 42.3 | Improved |
| Train timeline model on OF-Syn instead of OF-Staged (ablation) | ASQuery · OF-ItW | F1@50: 20.26 -> 27.60 | Improved |
| Combine OF-Staged and OF-Syn for timeline model (ablation) | ASQuery · OF-ItW | F1@50: 27.60 -> 28.61 | Improved |
Closest papers by meaning
- Kick Back & Relax: Learning to Reconstruct the World by Watching SlowTV (ICCV 2023 · similarity 0.84)
- SAFER-Activities: A Dataset for Smart Assessment of Fall Events and Routine Activities (ECCV 2026 · similarity 0.84)
- SceneScribe-1M: A Large-Scale Video Dataset with Comprehensive Geometric and Semantic Annotations (CVPR 2026 · similarity 0.83)
- Training-free Detection of Generated Videos via Spatial-Temporal Likelihoods (CVPR 2026 · similarity 0.83)
- PISA Experiments: Exploring Physics Post-Training for Video Diffusion Models by Watching Stuff Drop (ICML 2025 · similarity 0.82)
- Black Swan: Abductive and Defeasible Video Reasoning in Unpredictable Events (CVPR 2025 · similarity 0.82)
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.
