chenhsuanlin / bundle-adjusting-NeRF

BARF: Bundle-Adjusting Neural Radiance Fields 🤮 (ICCV 2021 oral)
MIT License
793 stars 114 forks source link

Documentation on the LLFF data scene? #4

Closed franciscoWizz closed 3 years ago

franciscoWizz commented 3 years ago

Hi Chen-Hsuan,

Do you have any documentation regarding how to set up another scene experiment? I did not found anything on the readme nor in the paper.

I'm refering to this folders and files: Data_LLFF

Thanks in advance,

chenhsuanlin commented 3 years ago

Hi @franciscoWizz, if you'd like to use your own sequence for BARF, you can create another data Python file (in data) to load your sequence. You can check out data/iphone.py as an example. If you want to use exactly the same data format as provided in LLFF scenes, please see the original NeRF repo for instructions.

franciscoWizz commented 3 years ago

Hi @chenhsuanlin,

Indeed I want to get the estimated camera poses out of a set of pictures. I'll take a look at the file you mention and come back if I have any question.

Thanks,

franciscoWizz commented 3 years ago

@chenhsuanlin,

When specifying the pictures dimensions I found that in the line #20 you assigned a particular size (1080, 1920):

image

2 questions about it:

  1. Does the model need that all the pictures be equally sized?
  2. Do I need to change the size read in line #20 with my images size? or this parameter is assigned in the command line options?

Thanks in advance

chenhsuanlin commented 3 years ago

Hi @franciscoWizz, for now we assume that the incoming frames from the sequence are all equally sized and specified in self.raw_H and self.raw_W. The yaml files specify how you would want to resize your images (opt.data.image_size), which would rescale your intrinsic parameters accordingly. Hope this helps!

chenhsuanlin commented 3 years ago

Closing for now, please feel free to reopen if you have further questions!