ActiveVisionLab/nerfmm

Train Configs for BLEFF

Closed this issue · 3 comments

HannahHaensen commented

Hey could you please share the configs for training Nerfmm on the Bleff Dataset?

Best Regards!

ziruiw-dev commented

Sorry about the delay. I have to dig into our old logs. We use this config for most BLEFF experiments:

alias='',
dir_enc_inc_in=True, 
dir_enc_levels=4, 
epoch=10000, 
eval_cam_interval=5, 
eval_img_interval=100, 
eval_img_num=1, 
eval_skip=1,
focal_lr=0.001, 
focal_lr_gamma=0.9, 
focal_milestones=range(0, 10000, 100), 
focal_order=2, 
fx_only=True, 
gpu_id=0, 
hidden_dims=128, 
learn_R=True, 
learn_focal=True, 
learn_t=True, 
multi_gpu=False, 
num_relu_epoch=0, 
num_rows_eval_img=10, 
num_sample=128, 
optimizer='Adam', 
pos_enc_inc_in=True, 
pos_enc_levels=10, 
pose_lr=0.001, 
pose_lr_gamma=0.9, 
pose_milestones=range(0, 10000, 100), 
rand_seed=17, 
res_dims=512, 
resize_ratio=2, 
scene_name='mix_rt/t005r005/bed1', 
sigmoid_density=False, 
start_refine_focal_epoch=-1, 
start_refine_pose_epoch=-1, 
static_lr=0.001, 
static_lr_gamma=0.9954, 
static_milestones=range(0, 10000, 10),
store_pose_history=True, 
train_img_num=-1, 
train_rand_cols=32, 
train_rand_rows=32, 
train_skip=1, 
true_rand=False, 
use_dir_enc=True, 
use_ndc=True, 
white_bkgd=False,
HannahHaensen commented

Thanks! @ziruiw-dev
another question is the refine_nerfmm file always used after training?