You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
When using 3dgan_mit_biasfree in test mode in order to make new objects, testGAN function for generating new objects, there is a line that thresholds the values at .5 to create voxels.
At first I thought maybe it does that because the values represent probabilities on [0,1] of being occupied so you just do a hard threshold.
But actually when plotting the distribution of the values, about half of them are below 0. They almost look like a laplace distribution or something. So what is supposed to be represented here by g_objects (32x64x64x64x1) is (batchsize x N x N x N x 1), and is thresholding > .5 what we want vs > 0?
The text was updated successfully, but these errors were encountered:
When using 3dgan_mit_biasfree in test mode in order to make new objects, testGAN function for generating new objects, there is a line that thresholds the values at .5 to create voxels.
At first I thought maybe it does that because the values represent probabilities on [0,1] of being occupied so you just do a hard threshold.
But actually when plotting the distribution of the values, about half of them are below 0. They almost look like a laplace distribution or something. So what is supposed to be represented here by g_objects (32x64x64x64x1) is (batchsize x N x N x N x 1), and is thresholding > .5 what we want vs > 0?
The text was updated successfully, but these errors were encountered: