FreemanG

I am confused by your synthetic dataset: 1. How do you generated the ground-truth frames for the 36 virtual cameras? 2. Why does the number of virtual cameras changes from...

How to ensure reproducibly data augmentation.

3

In my experiments, I noticed that the data augmentation was not deterministic between different runs, even with the numpy and tf random seeds both being set.

Poses are camera-to-world?

3

According to your data description: > Poses are camera-to-world, not world-to-camera transformations. Then why does [normalize_cam_dict.py](https://github.com/Kai-46/nerfplusplus/blob/master/colmap_runner/normalize_cam_dict.py) save W2C in the cam_dict? Shouldn't it be C2W (cam to world)?

Thanks for sharing this great work! Can dynibar be trained on the short Nvidia dataset that NSFF used (24 frames), or [DynamicNeRF](https://github.com/gaochen315/DynamicNeRF) used (12 frames).