Thank you team for the excellent work, I try to reproduce your results, but I only have 26 on psnr, my cofigs are as follows, do you have any suggestions to improve my on psnr? I see that you can reach 36 in your paper

[lll123-cpu]

My data shape is 512512D CT（Lung window cross-section） nii.gz

example config for CuNeRF

training

max_iter : 400000
eval_iter: 10000
save_iter: $eval_iter
log_iter : 100
bs_train : 384 # batch size
bs_eval : 4096
bs_test : $bs_eval
N_eval : 50
lr : 0.002
lr_final : 0.00002
workers : 4

modeling

stype : cube
radius : 1
n_samples : 64
n_imps : 64
mtype : NeRFMLP

test

zpos : [0]
scales : [2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0]
angles : [0]
asteps : 45
axis : [0, 1, 1]
cam_scale : 1.5

path

result_path : /home/h3c/lmt/code/CuNeRF-main/save/results
save_path : /home/h3c/lmt/code/CuNeRF-main/save/savepic
log_path : /home/h3c/lmt/code/CuNeRF-main/save/logs

loss :
name : Adaptive_MSE_LOSS

sampling :
stype : $stype
n_samples : $n_samples

rendering :
stype : $stype

importance :
stype : $stype
n_samples : $n_imps

model :
name : $mtype
netD : 8
netW : 256
in_ch : 3
out_ch : 2
skips : [4]
max_freq: 10
p_fns : ['sin', 'cos']

model_ft : $model

dataset :
mode : $mode
dname: Medical3D
train :
file : $file
scale : $scale
radius : $radius
bsize : $bs_train
modality: $modality

eval :
file : $file
scale : $scale
radius : $radius
N_eval : $N_eval
bsize : $bs_eval
modality: $modality

test :
file : $file
scales : $scales
radius : $radius
axis : $axis
angles : $angles
zpos : $zpos
cam_scale : $cam_scale
asteps : $asteps

dataloader :
mode : $mode
train :
batch_size : 1
shuffle : True

eval:
batch_size : 1
shuffle : False

test:
batch_size : 1
shuffle : False

optim :
name : Adam
lr : $lr
betas: [0.9, 0.999]
eps : 0.000001

metrics :
psnr : True
ssim : True
lpips : False
avg : False

lr_decay :
lr_init : $lr
lr_final : $lr_final
max_iter : $max_iter
lr_delay_steps : 1000
lr_delay_mult : 0.01

clip_grad :
max_val : 0.1
max_norm : 0.1

[djx99]

have you solved?

[djx99]

Loading model from: /home/mul/anaconda3/envs/lee_son_code/lib/python3.9/site-packages/lpips/weights/v0.1/alex.pth
max_iter : 200000
eval_iter : 10000
save_iter : 10000
log_iter : 100
bs_train : 1024
bs_eval : 32768
bs_test : 32768
N_eval : 50
lr : 0.002
lr_final : 2e-05
workers : 4
stype : cube
radius : 1
n_samples : 64
n_imps : 64
mtype : NeRFMLP
zpos : [0]
scales : [1]
angles : [0]
asteps : 45
axis : [0, 1, 1]
cam_scale : 1.5
result_path : save/CuNeRFx1/liver_choose_1-50
save_path : save/CuNeRFx1/liver_choose_1-50
log_path : save/CuNeRFx1/liver_choose_1-50
loss
name : Adaptive_MSE_LOSS

sampling
stype : cube
n_samples : 64

rendering
stype : cube

importance
stype : cube
n_samples : 64

model
name : NeRFMLP
netD : 8
netW : 256
in_ch : 3
out_ch : 2
skips : [4]
max_freq : 10
p_fns : ['sin', 'cos']

model_ft
name : NeRFMLP
netD : 8
netW : 256
in_ch : 3
out_ch : 2
skips : [4]
max_freq : 10
p_fns : ['sin', 'cos']

dataset
mode : train
dname : Medical3D
train : {'file': 'data/liver_choose_1-50.nii.gz', 'scale': 1, 'radius': 1, 'bsize': 1024, 'modality': None}
eval : {'file': 'data/liver_choose_1-50.nii.gz', 'scale': 1, 'radius': 1, 'N_eval': 50, 'bsize': 32768, 'modality': None}
test : {'file': 'data/liver_choose_1-50.nii.gz', 'scales': [1], 'radius': 1, 'axis': [0, 1, 1], 'angles': [0], 'zpos': [0], 'cam_scale': 1.5, 'asteps': 45}

dataloader
mode : train
train : {'batch_size': 1, 'shuffle': True}
eval : {'batch_size': 1, 'shuffle': False}
test : {'batch_size': 1, 'shuffle': False}

optim
name : Adam
lr : 0.002
betas : [0.9, 0.999]
eps : 1e-06

metrics
psnr : True
ssim : True
lpips : False
avg : False

lr_decay
lr_init : 0.002
lr_final : 2e-05
max_iter : 200000
lr_delay_steps : 1000
lr_delay_mult : 0.01

clip_grad
max_val : 0.1
max_norm : 0.1

expname : CuNeRFx1
cfg : configs/example.yaml
scale : 1
mode : train
file : data/liver_choose_1-50.nii.gz
save_map : True
resume : True
resume_type : current
is_details : True
is_gif : False
is_video : False
modality : None

Resuming from save/CuNeRFx1/liver_choose_1-50/current.pkl
Failing in resuming from save/CuNeRFx1/liver_choose_1-50/current.pkl
0%| | 1/200000 [00:00<?, ?it/s]
[TRAIN] Iter: 10000 LR : 0.001715 LOSS : 0.000503 PSNR : 33.935545: 5%|███▉ | 10000/200000 [05:01<1:35:29, 33.16it/s]Saving checkpoint at save/CuNeRFx1/liver_choose_1-50/current.pkl
[EVAL] : 49: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 50/50 [02:29<00:00, 3.00s/it]
[BEST] psnr : 26.20625455235359 | ssim : 0.8397008776664734 | lpips : 0.2347143143415451 | avg : 0.06083122525019683███████████████████████████████████████████████| 50/50 [02:29<00:00, 2.99s/it]
[EVAL] PSNR : 26.20625455235359 | SSIM : 0.8397008776664734 | LPIPS : 0.2347143143415451 | AVG : 0.06083122525019683
[TRAIN] Iter: 20000 LR : 0.001567 LOSS : 0.000294 PSNR : 35.973539: 10%|███████▉ | 20000/200000 [12:40<1:31:01, 32.96it/s]Saving checkpoint at save/CuNeRFx1/liver_choose_1-50/current.pkl
[EVAL] : 49: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 50/50 [02:30<00:00, 3.00s/it]
[BEST] psnr : 26.20625455235359 | ssim : 0.8587431907653809 | lpips : 0.2015404999256134 | avg : 0.06011395948083104███████████████████████████████████████████████| 50/50 [02:30<00:00, 3.00s/it]
[EVAL] PSNR : 25.424410251404193 | SSIM : 0.8587431907653809 | LPIPS : 0.2015404999256134 | AVG : 0.06011395948083104