Research map / Video / Video-language understanding
EgoPolice: A Benchmark for Egocentric Video Understanding in High-Stakes Police Body-Worn Camera Footage
Max Gonzalez Saez-Diez, Jihoon Chung, Adam D. Wolsky, Greg Lanzalotto, Dean Knox, Jonathan Mummolo, Brandon Stewart, Olga Russakovsky
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 |
|---|---|---|---|
| use video-pretrained VideoMAE V2 instead of image-pretrained CLIP | linear probe · 1min window, OL | mean F1: 46.1 -> 49.8 | Improved |
| use video-pretrained VideoMAE V2 instead of image-pretrained DINOv2 | linear probe · 1min window, ID | mean F1: 54.2 -> 58.4 | Improved |
| use VideoMAE V2 instead of CLIP on 1sec ID | linear probe · 1sec window, ID | mean F1: 39.2 -> 37.9 | No clear effect |
| use VideoMAE V2 instead of CLIP on Medical Treatment | linear probe · BWC Medical Treatment, ID 1s | F1: 40.0 -> 31.2 | Hurt |
| use VideoMAE V2 instead of CLIP on Running | linear probe · BWC Running, ID 1s | F1: 2.9 -> 15.0 | Improved |
| use VideoMAE V2 instead of CLIP on Handcuffing | linear probe · Handcuffing, ID 1s | F1: 4.1 -> 15.8 | Improved |
| use VideoMAE V2 instead of CLIP on Civilian Injured | linear probe · Civilian Injured, ID 1s | F1: 56.3 -> 39.7 | Hurt |
| lengthen clip from 1sec to 10sec | Qwen2.5-VL / VLMs · zero-shot MCQ | accuracy: 52.5 -> 55.8 | Improved |
| lengthen clip from 10sec to 1min | Gemini 2.5 Flash · zero-shot MCQ | accuracy: 69.7 -> 66.1 | Hurt |
| lengthen window from 1sec to 1min | linear probe VideoMAE V2 · ID | mean F1: 37.9 -> 58.4 | Improved |
| use Gemini 2.5 Pro on 1min clips versus random | Gemini 2.5 Pro · zero-shot MCQ 1min | accuracy: 20.0 -> 76.9 | Improved |
| small proprietary model GPT-4.1-Nano | GPT-4.1-Nano · zero-shot MCQ 1min | accuracy: 20.0 -> 10.8 | Hurt |
Closest papers by meaning
- Action Scene Graphs for Long-Form Understanding of Egocentric Videos (CVPR 2024 · similarity 0.87)
- Ego-Only: Egocentric Action Detection without Exocentric Transferring (ICCV 2023 · similarity 0.87)
- Actor and Observer: Joint Modeling of First and Third-Person Videos (CVPR 2018 · similarity 0.87)
- EgoSim: An Egocentric Multi-view Simulator and Real Dataset for Body-worn Cameras during Motion and Activity (NeurIPS 2024 · similarity 0.85)
- EgoExoBench: A Benchmark for First- and Third-person View Video Understanding in MLLMs (NeurIPS 2025 · similarity 0.85)
- Retrieval-Augmented Egocentric Video Captioning (CVPR 2024 · 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.
