Benchmark RadarK9-Bench
K9-Bench is a benchmark focused on real-world domestic dog videos, with approximately 5,000 question-answer pairs across 907 videos spanning 5 task categories. The tasks test long-form, canine-centric multimodal reasoning, including action and interaction understanding. A VLM/LLM-powered pipeline was used for data generation.
- Released
- 2026-07-02
- Readiness
- Inspectable
- Primary field
- Robotics & Autonomous Systems
Why it matters
It addresses the underexplored application of multimodal models to animal-centric scenarios, providing a benchmark to evaluate models on recognizing distress signals and enabling responsive robotic companions.
Motivation
MLLMs have shown strong zero-shot capabilities across diverse inputs such as across images, video, audio, and text.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.