search

zju3dv / NeuralRecon

Code for "NeuralRecon: Real-Time Coherent 3D Reconstruction from Monocular Video", CVPR 2021 oral
https://zju3dv.github.io/neuralrecon/
Apache License 2.0
2.03k stars 294 forks source link

one question about occ_vol and tsdf_vol #54

Open HeCraneChen opened 2 years ago

HeCraneChen commented 2 years ago

Thanks for the nice code! One quick question about implementation detail in datasets/transforms.py. I noticed that the occ_vol is set to be a subset of tsdf_vol obtained from nn.functional.grid_sample (as shown in the figure below). However, tsdf_vol can also be calculated when calling tsdf_vol, weight_vol = tsdf_vol.get_volume() in line 302. But you didn't not use that tsdf_vol. Instead, lines 318~327 calculated tsdf_vol in another way using the tsdf generated from meta data. I wonder, is there any special concern for doing this? Thanks!

image
blacklioneer commented 2 years ago

may I ask some question about data['vol_origin'] in transform.py. Why this code didn't do random transform to this parameter since the author used it on computation with transformed volume in Line264