render error

[yeyan00]

i user small city data,set conf use matrixcity_2block_lowquality.txt,just modify data_dir， the train process is ok, but render error as follow:

#cmd like this
app/renderer.py --config confs/matrixcity/matrixcity_2block_lowquality.txt --ckpt= nerf/LandMark/log/matrix_city_block_1+2_multi/matrix_city_block_1+2_multi.th 

#error
size mismatch for basis_mat.weight: copying a param with shape torch.Size([27, 144]) from checkpoint, 
the shape in current model is torch.Size([27, 48]).
size mismatch for basis_den.weight: copying a param with shape torch.Size([1, 48]) from checkpoint, 
the shape in current model is torch.Size([1, 16]).

[Saaaaln]

We used the following commands for training and rendering and do not encounter the problem you mentioned.

cd ./LandMark
 # train
python app/trainer.py --config confs/matrixcity/matrixcity_2block_lowquality.txt
 # render
python app/renderer.py --config confs/matrixcity/matrixcity_2block_lowquality.txt --ckpt log/matrix_city_block_1+2_lowquality/matrix_city_block_1+2_lowquality.th`

Could you provide the detailed training and rendering config?

[yeyan00]

We used the following commands for training and rendering and do not encounter the problem you mentioned.

cd ./LandMark
 # train
python app/trainer.py --config confs/matrixcity/matrixcity_2block_lowquality.txt
 # render
python app/renderer.py --config confs/matrixcity/matrixcity_2block_lowquality.txt --ckpt log/matrix_city_block_1+2_lowquality/matrix_city_block_1+2_lowquality.th`

Could you provide the detailed training and rendering config?

i use same config training and render, just modify data_root

dataroot = ./nerf/LandMark/data/matrixcity/bdaibdai___MatrixCity/small_city/aerial/pose
datadir = block_A
expname = matrix_city_block_1+2_lowquality
dataset_name = matrixcity
ndims = 1

nonlinear_density=True

lb=[-1.0378e+01, -7.4430e+00, -1.1921e-07]
ub=[1.2078e+00,  2.4078e+00,  9.0000e-01]

add_nerf = 5000

basedir = ./log

train_near_far=[0.6, 3.5]

downsample_train = 1

batch_size = 8192

N_voxel_init  = 2097156    # 128**3
N_voxel_final = 27000000  # 300**3

upsamp_list = [2000,3000,4000,5500,7000]
update_AlphaMask_list = [2000,4000,8000,10000]

lr_decay_iters = 8000
n_iters = 8000

N_vis = 5
vis_every = 5000

n_lamb_sigma = [16,16,16]
n_lamb_sh    = [48,48,48]

fea2denseAct = relu
shadingMode = MLP_Fea

view_pe = 2
fea_pe = 2

L1_weight_inital = 8e-5
L1_weight_rest = 4e-5
rm_weight_mask_thre = 1e-4

TV_weight_density = 0.1
TV_weight_app = 0.01

compute_extra_metrics = 1
run_nerf = 0
bias_enable = 1
white_bkgd = 1
sampling_opt = 0

[Saaaaln]

From this error, the config you use for training and rendering may be not the same.

We have provided the download link of demo ckpt in the latest release.

The links and usage are introduced in README.