how to generate cfg_args.json and ntc_parms for our own dataset?

[AnudhinaDhanabalan]

I am currently working to generayte 3dgstream for my own dataset, during the final step i am confused abt getting two things

[AnudhinaDhanabalan]

i) the cfg_args.json file inside the flame_steak_init, how to generate that for my own dataset? ii) under the ntc folder there is a file named ntc_path_F_4.pth , it has been mentioned in the cfg_args.json file ,so if want my own file of that what should i do? thanks in advance !!

[SJoJoK]

1. The cfg_args.json file is automatically generated based on the arguments you set during training. You need to adjust the arguments according to your dataset, and once training starts, the cfg_args.json file will be created in the output directory.

2. To create your own config of ntc, please refer to the documentation of tiny-cuda-nn.

[AnudhinaDhanabalan]

python train_frames.py --read_config --config_path /METACHN/custom/custom_init/cfg_args.json -o /METACHN/custom/customoutput -m /METACHN/custom/custom_init -v /METACHN/custom_ff2 --image images --first_load_iteration 15000 --quiet ,but it needs and asks for a cfg_args.json file if i run the above code

[SJoJoK]

you dont need to specify --read_config

[AnudhinaDhanabalan]

it is working, but i am getting this error now !! assert len(extra_f_names)==3*(self.max_sh_degree + 1) ** 2 - 3 AssertionError

[SJoJoK]

Could you please read the README , our paper or just take a look at the provided cfg_args.json and try to understand why these error raise but not directly ask them in the issue? I think it's clear that you should specify --sh_degree 1, and this is described in our paper, in our README, and in the provided cfg_args.json.

[AnudhinaDhanabalan]

okayy, thank you so much !