sensefi

Description

SenseFi (Yang et al., NTU) is a benchmark library for deep-learning-based WiFi CSI sensing. It bundles standardised dataloaders, model implementations, and the ntu-fi CSI datasets so that methods for activity recognition, gesture recognition, and human identification can be compared on a common footing. SenseFi acts both as a code library and as a reference dataset suite in the CSI sensing community.

Modality / size

• Modality: WiFi CSI (commodity NICs).
• Datasets bundled: includes the NTU-Fi HAR and NTU-Fi Human-ID subsets and adapters for several public CSI datasets.
• Models bundled: reference implementations of MLP, CNN, RNN, ViT, and other CSI baselines.

Used by (papers)

• Standard reference for CSI deep-learning baselines in the vault.
• Cited by CSI surveys and by methods that report numbers on NTU-Fi.

Notes

• Open-source release on GitHub; companion paper from NTU.
• See ntu-fi for the dataset subsets shipped alongside SenseFi.