SIS-Bench is a comprehensive benchmark constructed from three diverse UAV video sources for evaluating multimodal large language models. It assesses spatial cognition and self-awareness across three cognitive levels of increasing difficulty: perception, memory, and reasoning. The benchmark was created by choucsan and last updated on July 13, 2026.
Use Cases
- Benchmarking model spatial cognition based on UAV video perception tasks.
- Evaluating model self-awareness based on memory and reasoning challenges.
- Training multimodal models for embodied UAV scenario understanding.
Strengths
- Assesses models along two complementary dimensions: spatial cognition and self-awareness.
- Built from three diverse UAV video sources: AirScape, UrbanVideo-Bench, and VisDrone.
- Evaluates across three cognitive levels of increasing difficulty.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Dataset size and row count are unknown, which may limit suitability assessment.
- Freshness should be verified; last updated 2026-07-13.
Provenance
- Source
- huggingface
- Collection Method
- Constructed from three UAV video sources.
- Freshness
- Last updated 2026-07-13.