Bo Yang
Bo Yang
@casawa The array (144, 254, 67, 1) is a voxel grid (x, y, z, channel), the value of the channel is either 1 or 0, indicating occupied or not. This...
hi @caixialiu-bjut, if you need to compute the IoU of 32^3 or 64^3, you can downsample the predicted 256^3 voxels to 32^3 or 64^3. The intermediate layers are features, not...
Hi @samomin46 , in the paper, I use matlab to visualize the 3D voxel. Here's the code. Hope it's helpful for your research. function mesh(vol) threshold=0.5; [d1,d2,d3]=size(vol); figure; [x,y,z]=meshgrid(1:1:d1,1:1:d2,1:1:d3); p1=patch(isosurface(x,y,z,vol,threshold,'noshare'));...
Hi @Hans1984 , according to the definition, the loss is negative. It indeed takes some time to converge at the beginning, you are suggested to track it over a few...
@kunalkhadilkar sorry for the delay. which version of tf do you use? It works fine on my pc. The released trained model should be able to take in the dynamic...
Hi @254788 , make sure the virtual environment is correct before running the code. To speed up the code, you are suggested to modify the latent vector to be a...
@yzp12 Yes, the output of the discriminator is a vector instead of a single scalar. Mean value of that vector is the loss. For a real reconstruction pair, the mean...
Hi @bharadwajdhornala , the numpy viewer is indeed bad. You are suggested to use Matlab instead.
Hi @feidasheng, you need to run the pre-trained model to generate the demo result.
hi @Hans1984 , thanks for your interest in our paper. The depth map resolution is 640x480.