search

chenhsuanlin / bundle-adjusting-NeRF

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

A question #48

Closed lllz815 closed 2 years ago

lllz815 commented 2 years ago

after i train a barf and a image align model ,i get a .ckpt file ,how i can use the model or test the model

chenhsuanlin commented 2 years ago

Hi @lllz815, you should run python3 evaluate.py [--args]; please see the README for details.

lllz815 commented 2 years ago

Hi @lllz815, you should run python3 evaluate.py [--args]; please see the README for details.

thank for your replying ,now i train a barf with some picture which is taked by my ( xiao mi) phone ,i use the iphone.py and i do'nt know how to compute th focal ,so use raw_W*4.2/(12.8/2.55) as you ,but i finally get a bad result i use the follow commend to train python train.py --group=barf --model=barf --yaml=barf_iphone --name=iphone2 --data.scene=chair --arch.posenc! and the follow commend to evaluate python evaluate.py --group=barf --model=barf --yaml=barf_iphone --name=iphone2 --data.scene=chair --resume=10000 --arch.posenc! my images are like this
14 and after evaluate get the result depth_14 rgb_14

for this result ,canyou give me some suggesttions ,thanks very much!!

chenhsuanlin commented 2 years ago

Are you trying to capture the chair with an object-centric sequence? If so, you may want to check out some previous discussions (e.g. #19, #32). BARF will probably be unable to automagically find the camera poses from scratch if the scenes were not simple forward-facing; you may need to run COLMAP to find a good pose initialization first.