Research map / Video / Video-language understanding
PriVi: Towards a General-Purpose Video Model for Primate Behavior in the Wild
Felix B. Mueller, Jan F. Meier, Timo Lueddecke, Richard Vogg, Roger L. Freixanet, Valentin Hassler, Tiffany Bosshard, Elif Karakoc, William J. O'Hearn, Sofia M. Pereira, Sandro Sehner, Kaja Wierucka and 11 more
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 |
|---|---|---|---|
| continue V-JEPA pretraining on PriVi instead of human-centric data only | V-JEPA ViT-L + attentive classifier · ChimpACT (test) | mAP: 36.33 -> 39.25 | Improved |
| continue V-JEPA pretraining on PriVi instead of human-centric data only | V-JEPA ViT-L + attentive classifier · BaboonLand (test) | accuracy: 74.91 -> 75.43 | Improved |
| add dataset-level pretraining on top of PriVi (ablation) | PriVi V-JEPA · PanAf500 (test) | accuracy: 86.74 -> 85.01 (balanced acc 62.75 -> 62.96) | Mixed |
| add dataset-level pretraining on top of PriVi (ablation) | PriVi V-JEPA · ChimpBehave (test) | balanced accuracy: 71.30 -> 75.14 | Improved |
| use relevance-filtered YouTube instead of random YouTube (ablation) | V-JEPA pretraining · ChimpACT (val) | mAP: 33.87 -> 37.88 | Improved |
| add R&O to YT-Filtered (PriVi) (ablation) | V-JEPA pretraining · PanAf500 (val) | balanced accuracy: 76.33 -> 79.95 | Improved |
| domain-level PriVi vs dataset-level pretraining on target only (ablation) | V-JEPA pretraining · ChimpACT (val) | mAP: 35.86 -> 38.75 | Improved |
| add dataset-level pretraining on PanAf500 only vs PriVi (ablation) | V-JEPA pretraining · PanAf500 (val) | balanced accuracy: 79.95 -> 78.57 | Hurt |
| add PriVi + DaLP on other target dataset's data (ablation) | PriVi V-JEPA · PanAf500 (val) | balanced accuracy: 79.95 -> 69.05 | Hurt |
| add primate-centric cropping to pretraining (ablation) | PriVi V-JEPA · ChimpACT (val) | mAP: 32.62 -> 38.75 | Improved |
| add linear downprojection to classifier (ablation) | attentive classifier · PanAf500 (val) | balanced accuracy: 62.52 -> 79.95 | Improved |
| use three classifier layers instead of one (ablation) | attentive classifier · ChimpACT (val) | mAP: 35.68 -> 38.75 | Improved |
Closest papers by meaning
- The PanAf-FGBG Dataset: Understanding the Impact of Backgrounds in Wildlife Behaviour Recognition (CVPR 2025 · similarity 0.90)
- Animal-Bench: Benchmarking Multimodal Video Models for Animal-centric Video Understanding (NeurIPS 2024 · similarity 0.88)
- MammAlps: A Multi-view Video Behavior Monitoring Dataset of Wild Mammals in the Swiss Alps (CVPR 2025 · similarity 0.86)
- BEAST3D: animal behavioral analysis and neural encoding from multi-view video via Gaussian splatting (NeurIPS 2026 · similarity 0.86)
- Generating the Wild: Individual-Consistent Image-to-Video Generation for Wildlife (NeurIPS 2026 · similarity 0.86)
- BuckTales: A multi-UAV dataset for multi-object tracking and re-identification of wild antelopes (NeurIPS 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.
