Right now, it looks like Edge Impulse only works with Time Series Data as the only 1D input block.
However, I know that I can use any data (image, sound, motion, etc.), and it does not have to be a time series. If I add flattened image data and use it as raw input to a DNN, the feature extraction block gives me errors, as the image data size does not match the “expected window size.”
I recommend having a “raw” input block that lets me just pass my pre-configured data all the way through to the learning block. It feels a bit “hacky” that I need to set my Input Block window size to match my data sample size (784 x 1 matrix). I had to assume that the sampling rate was 1 ms when sending data to my project (even though there is no sample rate).