training pipeline for my own data

[flugenheimer]

Hello,

I am a big fan of neural radiance fields and i am currently experimenting with it for more industrial purposes. I think that you are solving a very relevant problem!

Would it be possible to include a description of a training pipeline from having the captured images to running your code?

[bennyguo]

Sorry for the late reply!

First use the imgs2poses.py script from the LLFF repo to genereate camera parameters used for training (the poses_bound.npy file and the sparse folder). Then you have to create a train.txt and a val.txt containing training and validation images. The refl_masks folder is optional and I manually paint the masks for some of the images using PhotoShop. Finally organize all the files according to the structures of the RFFR dataset and you are good to go!

I'll add this data generation pipeline to README.