search

NVlabs / nvdiffrec

Official code for the CVPR 2022 (oral) paper "Extracting Triangular 3D Models, Materials, and Lighting From Images".
Other
2.13k stars 223 forks source link

what's the poses_bounds.npy's data format #147

Closed 2226415223 closed 11 months ago

2226415223 commented 11 months ago

what's the poses_bounds.npy's format, I'm using image sequence instead of original obj to rebuild mesh, I know the camera parameters, how to fill to the poses_bounds.npy.

JHnvidia commented 11 months ago

Hi,

It's the format used in "Local Light Field Fusion". There's some brief documentation at their github page: https://github.com/Fyusion/LLFF

Depending on how your data is stored you could also consider writing your own dataloader. You can see the expeced output here: https://github.com/NVlabs/nvdiffrec/blob/main/dataset/dataset_llff.py#L102

2226415223 commented 11 months ago

Thanks, I'm trapped in analysing the format via inverse the numpy operation, you remind me why not set the camera parameters directly!

JHnvidia commented 11 months ago

Our code uses an OpenGL like format with 4x4 model-view-projection matrices to specify the camera (as per the output of the dataset).

Datasets based on rendering packages, such as used in NeRF (using blender) typically dump the matrix straight away: https://github.com/NVlabs/nvdiffrec/blob/main/dataset/dataset_nerf.py#L66.

The LLFF dataloader is just included because we wanted to use some datasets from previous work. The camera parameters are kind of messy.