Issue with the generation of EAsparse_{idx}.npz files

[happiness2b]

Based on the file generate_4d_polar_percentile.py, the EAsparse_{idx}.npz files generated in the /radar_tensor_percentile folder seem to select the top 10% of signals with the highest energy from the REA tensor, rather than extracting the top-Nr strongest signals from each range segment. Could you please confirm whether my understanding is correct?

Thank you in advance for your help.

[Toytiny]

if you didn't see any slices upon the range axis, it would mean 'selecting the top 10% singles from the whole REA tensor'.

[Ericwen2001]

Sorry for the confusion and point out this. I checked the code and confirmed that we are using generate_4d_polar_doppler.py, not generate_4d_polar_percentile.py (which is only used for the ablation study in the paper).

The script loads the 4D data and computes the mean along the Doppler axis to obtain a 3D spatial power cube of shape [R, A, E].

For each range bin, it selects the top 250 directions (in azimuth-elevation space) with the highest power, and maps them back to their corresponding (range, azimuth, elevation) coordinates.

For each selected spatial point, it retrieves the full Doppler profile (i.e., power values across all Doppler bins), and then:

Extracts the top 3 Doppler bins and their corresponding power values,

Computes the mean and variance of the full Doppler spectrum.

These statistics are stacked into an array of shape [8, N], where N is the total number of selected spatial points. This array contains:

Top 3 Doppler power values

Their Doppler bin indices

The mean and variance of the Doppler values.

Hope this can answer your question. @happiness2b