Research map / Video / Video-language understanding
EthoCLIP: Ontology-Enhanced Video-Language Pretraining for Animal Behavior Understanding
Yinuo Jing, Jinyan Wu, Zixi Yang, Kongming Liang, Xiatian Zhu, Zhanyu Ma
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 ontology semantic embedding (labels+synonyms+definitions) (ablation) | X-CLIP · AnimalBand test set (behavior classification) | overall mAP: 61.20 -> 62.54 | Improved |
| add ontology-aware graph modeling (ablation) | X-CLIP · AnimalBand test set (behavior classification) | overall mAP: 61.20 -> 62.25 | Improved |
| combine OSE and OAGM (ablation) | EthoCLIP · AnimalBand test set (behavior classification) | overall mAP: 62.54 -> 62.98 | Improved |
| use OSE weighted embedding instead of concatenation (ablation) | EthoCLIP · AnimalBand test set (behavior classification) | overall mAP: 61.40 -> 62.54 | Improved |
| use OSE instead of LLM keyword extraction (ablation) | EthoCLIP · AnimalBand test set (behavior classification) | overall mAP: 61.92 -> 62.54 | Improved |
| use 2 hops instead of 1 or 3 in graph attention (ablation) | EthoCLIP · AnimalBand test set (behavior classification) | overall mAP: 61.93 -> 62.98 (1 hop), 62.23 (3 hops) | Improved |
| pretrain on AnimalBand instead of directly merged labels (ablation) | X-CLIP · Transfer mean (Panda, SheepActivity, ChimpACT) | mean score: 69.04 -> 74.26 | Improved |
| pretrain on AnimalBand instead of Kinetics-400 (ablation) | X-CLIP · Transfer mean | mean score: 64.22 -> 74.26 | Improved |
| use ontology-enhanced EthoCLIP over X-CLIP | AnimalBand-pretrained · Transfer mean | mean score: 74.26 -> 78.11 | Improved |
| partially unfreeze temporal transformer and prompting modules (ablation) | EthoCLIP · Transfer mean | mean score: 78.11 -> 87.72 | Improved |
| use EthoCLIP over X-CLIP on tail labels | EthoCLIP · AnimalBand test set (behavior classification) | tail mAP: 50.36 -> 53.27 | Improved |
| use 40% of AnimalBand pretraining data (ablation) | EthoCLIP · Panda (transfer) | Acc@1: 84.80 -> 72.00 (40%) | Mixed |
Closest papers by meaning
- Meow-Omni 1: A Multimodal Large Language Model for Feline Ethology (NeurIPS 2026 · similarity 0.87)
- Animal-Bench: Benchmarking Multimodal Video Models for Animal-centric Video Understanding (NeurIPS 2024 · similarity 0.87)
- UniAP: Towards Universal Animal Perception in Vision via Few-Shot Learning (AAAI 2024 · similarity 0.86)
- BioCLIP 2: Emergent Properties from Scaling Hierarchical Contrastive Learning (NeurIPS 2025 · similarity 0.85)
- AniMo: Species-Aware Model for Text-Driven Animal Motion Generation (CVPR 2025 · similarity 0.84)
- MammAlps: A Multi-view Video Behavior Monitoring Dataset of Wild Mammals in the Swiss Alps (CVPR 2025 · similarity 0.84)
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.
