ntu-fi

Description

NTU-Fi is a family of WiFi CSI datasets released alongside the SenseFi benchmark library (NTU). It contains two main subsets: NTU-Fi HAR for human activity recognition and NTU-Fi Human-ID for identification of subjects from gait-induced CSI patterns. The two subsets share collection setup so they are commonly used as a transfer-learning pair (HAR -> Human-ID).

Modality / size

• Modality: WiFi CSI on commodity 5 GHz hardware.
• Subjects / scenarios: multiple subjects, indoor lab.
• Labels: activity classes (HAR subset) and person identity (Human-ID subset).
• Exact sample counts: see SenseFi paper / repository.

Used by (papers)

• CSI feature-extractor transfer studies in the vault use NTU-Fi HAR -> NTU-Fi Human-ID to demonstrate that features learned for activity recognition transfer to person identification.
• Default benchmark in the SenseFi reference implementation.

Notes

• Released as part of the sensefi benchmark library.
• HAR and Human-ID subsets are folded into this single note because they share collection methodology and are almost always cited together.