finds.dev← search

// the find

Sentdex/BCI

★ 349 · Python · MIT · updated Oct 2022

Brain-Computer interface stuff

A hobbyist BCI project from Sentdex using an OpenBCI 16-channel EEG headset to classify imagined left/right movements via FFT features fed into a TensorFlow model. Best accuracy is ~63% on a 3-class problem (left/right/none), which is barely above chance for a balanced dataset. For developers curious about EEG signal processing and wanting real recorded data to experiment with without buying hardware.

Includes actual labeled EEG data (numpy arrays, ~10s clips, 16 channels, 60Hz FFT) that you can download and train on without owning the headset. Data format is straightforward — shape (250, 16, 60) with no custom loading code needed. Two pretrained SavedModel checkpoints are included so you can immediately run inference and compare your own models. The honest confusion matrix reporting is refreshing — the author doesn't dress up 63% accuracy as a win.

Last commit is October 2022 and the project was clearly abandoned mid-experiment — the stated goal of controlling GTA V never materialized. 63% accuracy on a 3-class problem with highly unbalanced difficulty means the model is not much better than guessing 'right' every time. No preprocessing pipeline is documented; the FFT step happens inside the OpenBCI GUI's networking tab, so reproducing the data collection setup requires that specific GUI configuration. External data download link (hkinsley.com) has no guarantee of permanence and the repo gives you no fallback.

View on GitHub →

// want more like this?

We dig through GitHub every week and send a few repos picked for what you actually care about — each with an honest take like this one.

Get finds in your inbox → Search again →