mmWave-QA: First Benchmark for Language-Conditioned mmWave Human Perception
A new arXiv paper introduces mmWave-QA, the first benchmark for language-conditioned millimeter-wave (mmWave) human perception. The research addresses the integration of large language models (LLMs) with mmWave sensing, a modality advantageous under low light and occlusion but previously unexplored due to scarce radar-language pairs, cross-dataset heterogeneity, and lack of a foundational mmWave encoder. The authors propose a minimal textualization interface that serializes mmWave point clouds into natural language, enabling off-the-shelf LLMs to operate in a question-answering (QA) setting. The benchmark aggregates data to evaluate LLMs' ability to understand mmWave radar data. The paper is available on arXiv under the identifier 2608.14179.
Key facts
- The paper is titled 'Can Language Models Understand mmWave Data? Benchmarking Large Language Models for mmWave Radar-Based Human Understanding'.
- The arXiv identifier is 2608.14179.
- The paper introduces mmWave-QA, the first benchmark for language-conditioned mmWave human perception.
- The research addresses the integration of LLMs with mmWave sensing.
- The approach uses a minimal textualization interface to convert mmWave point clouds into natural language.
- The benchmark enables LLMs to operate in a question-answering setting.
- The paper highlights challenges: scarcity of radar-language pairs, cross-dataset heterogeneity, and absence of a foundational mmWave encoder.
- The paper is published on arXiv, a preprint server.
- The announcement type is 'new'.
- The abstract mentions the advantages of mmWave under low light and occlusion.
Entities
Institutions
- arXiv