DietNerf-Demo / demo /src /config.py
alexlau's picture
first deploy demo
19677a1
raw
history blame
1.08 kB
# for downloading model from google drive
FILE_ID = "1iytA1n2z4go3uVCwE__vIKouTKyIDjEq"
MODEL_DIR = './models'
MODEL_NAME = 'trained_model'
class NerfConfig:
# MODEL CONFIG
model = "nerf"
net_activation = "relu"
rgb_activation = "sigmoid"
sigma_activation = "relu"
min_deg_point = 0
max_deg_point = 10
deg_view = 4
# reduce num_coarse_samples, num_fine_samples for speedup
num_coarse_samples = 32
num_fine_samples = 64
use_viewdirs = True
near = 2
far = 6
noise_std = None
# TODO @Alex: set white_bkgd as flag if we add LLFF dataset
white_bkgd = True
net_depth = 8
net_width = 256
net_depth_condition = 1
net_width_condition = 128
skip_layer = 4
num_rgb_channels = 3
num_sigma_channels = 1
lindisp = True
legacy_posenc_order = False
randomized = True
# DATA CONFIG
W = 800
H = 800
IMAGE_SHAPE = (W, H, 3)
# TODO @Alex: flexible focal if we add LLFF dataset
FOCAL = 555.5555155968841
# reduce CHUNK if OOM
CHUNK = 4096
DOWNSAMPLE = 2