Dataset Format / Pointcloud
weders opened this issue · 2 comments
Hi,
Thanks a lot for releasing this code!
When trying to get it running on my own data, I realized that the data preparation instructions are a bit unclear. If I have a dataset in the LLFF format. Do I need to transform it into the COLMAP format?
Further, is there a way to use the voxel embeddings without knowing the point cloud of the scene a priori?
Thanks!
Hi, weders.
Sorry that we haven't supported LLFF data format. But you can prepare the dataset in the form of nerf_synthetic (e.g., with camera poses, frame indices and image paths defined in a transform.json as our ToyDesk dataset).
Note that the LLFF dataloader usually comes with automatically pose centralization and near/far plane computation, while our dataloader requires manually assigning scale_factor, scene_center and near, far in the config file.
For the scene without point clouds, you can disable the use_voxel_embedding option at first, which might slightly decrease rendering details but still works fine with object-center learning.