Benchmark RadarAI BENCHMARK PROFILE
Prompting-MammAlps
Prompting-MammAlps is a camera-trap text-to-video retrieval benchmark, evaluating video-language models on fine-grained retrieval of ecological events. It includes a test set of 135 queries and 775 candidate videos, with a proposed method that combines action localization with LLM-based parsing.
- Released
- 2026-07-10
- Readiness
- Inspectable
- Primary field
- General AI
Why it matters
Text-to-video retrieval in ecological domains requires spatiotemporal understanding that current VLMs lack. This benchmark provides a standardized evaluation for fine-grained and interpretable retrieval, highlighting the limitations of zero-shot VLMs.
Motivation
Automatically retrieving videos from large camera-trap datasets remains challenging.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.