Must be sparse before gdc？

[mihudaner]

Hello, I am very interested in your article and code, and I have read the code carefully,

But I tried to use the depth map generated by the 64-line ground truth to correct the depth estimation result with gdc, and then down-sampling, but gdc has no effect

I must first convert the depth estimation result into pseudo-radar, then down-sampling it to 64 lines and then converting it into depth map, so that gdc can get the correct result

Do you have to convert the depth map to pseudo-radar, then sample the pseudo-radar, and then convert the depth map? It feels so cumbersome. Can't we use gdc for depth map first and then sample it down?

[Xi-Gong]

I see the --disable_subsample param under main_batch.py, have you ever try it?