N-WEAR events (DVS128) - event-camera version of the WEAR dataset
=====================================================================

Simulated DVS 128x128 event streams of the egocentric WEAR videos, as used in

  F. Sholichin et al., "N-WEAR: Spiking Event-Inertial Fusion with Hopfield Pooling
  for Energy-Efficient Egocentric Activity Recognition".
  Code: https://github.com/fauzisho/SNN-Fusion-HopfieldPooling

Contents: 18 subjects (WEAR sbj_0 ... sbj_17), one uncompressed zip per subject.
Each zip unpacks to

  sbj_<k>_events_dvs128/<activity>_part_<n>_from_<start>/<...>_<iii>_dvs128.h5
  sbj_<k>_events_dvs128/<activity>_part_<n>_from_<start>/v2e-args.txt

* one folder per annotated activity segment, cut into consecutive 2 s clips
* label = folder name before "_part_" (18 WEAR activities); NULL is not included
* H5: dataset "events", shape (N, 4), uint32, columns (t_us, x, y, p), x,y in 0..127,
  p in {0, 1}
* conversion: letterbox to 128x128, Super SloMo upsampling, v2e (bright illumination,
  pos/neg threshold 0.2, sigma 0.03, cutoff 200 Hz, leak 0.1 Hz, shot noise 1 Hz);
  exact arguments in each v2e-args.txt

Source video per subject: https://ubi29.informatik.uni-siegen.de/wear_dataset/raw/camera/sbj_<k>.mp4
Annotations: https://ubi29.informatik.uni-siegen.de/wear_dataset/annotations/60fps/wear_split_<k+1>.json
(every wear_split_*.json lists the segments of all subjects; split <k+1> holds out sbj_k)
Clip start times in the folder names of sbj_0 ... sbj_15 follow the 30 fps annotation
release, which differs from the 60 fps one by at most 0.04 s (same segments and labels).

Subject 2 has 17 activities: WEAR has no "jogging (rotating arms)" for this subject.

Download: https://thesiswearsnn.nbg1.your-objectstorage.com/n-wear-events-dvs128/index.html  (or data_preparation/download_events.sh in the repository)
Verify:   sha256sum -c SHA256SUMS

Licence: CC BY-NC-SA 4.0. This is a derivative of WEAR (Bock et al., IMWUT 2024,
https://mariusbock.github.io/wear/), which is licensed CC BY-NC-SA 4.0. Cite both WEAR
and the N-WEAR paper when using the data. Non-commercial use only.