How to train custom dataset

[s1752729916]

Hi.
Thank you for your work.

What should I prepare for training my custom dataset?

I have converted my dataset to the mega_nerf format and used meta_nerf style commands(building.yaml) to generate chunks, but the results is so bad and i think i miss some important steps when i prepare custom dataset. Do you have any scripts or tips for preparing custom dataset?

[MiZhenxing]

Hi, I use the same format to generate my custom dataset as Mega-NeRF. I mainly follow the instructions in mega-nerf. The linke is here https://github.com/cmusatyalab/mega-nerf#custom-data

The pipeline is typically as follows:

1. Run feature_extractor vocab_tree_matcher mapper in colmap to get the sparse reconstruction.
2. Run model_aligner in colmap to align the coordinates to the GPS information in the image
3. use mega_nerf/scripts/colmap_to_mega_nerf.py to convert from colmap to Mega-NeRF dataset.

You can refer to the document of colmap to learn how to run the programs. A possible problem of colmap_to_mega_nerf.py is that the ray_altitude_range in mage-nerf dataset is for the first axis i.e. x of the scene. So you need to make sure the x in your dataset is in the altitude direction when converting to mega-nerf. You can visualize the sparse points of colmap to see the axis.