def get_nfor_img_and_get_relmse(in_pth):
"""
!! 임시 !!
기존에 tungsten 랜더러에서 나온 buffer에서 nfor을 가져옴.
그리고 relMSE를 나오게 뽑음.
"""
input_spp_list = [32, 100, 256, 512, 1024, 2048, 4096] # [32, 100, 256, 512, 1024]
out_pth = in_pth + "/nfor"
if not os.path.exists(out_pth):
os.mkdir(out_pth)
f = open(out_pth + '/nfor_relMSE.txt', 'w')
ref_buffer = exr.read_all(os.path.join(in_pth, "out_64kspp.exr"))
ref_color = ref_buffer['diffuse'] + ref_buffer['specular']
for i in range(len(input_spp_list)):
input_name = "out_" + str(input_spp_list[i]) + "spp.exr"
input_buffer = exr.read_all(os.path.join(in_pth, input_name))
input_nfor = input_buffer['nfor']
rmse = calcRelMSE(input_nfor, ref_color)
rmse_str = str(input_spp_list[i]) + "spp image relMSE : " + str(rmse)
f.write(rmse_str)
f.write("\n")
print(rmse_str)
exr.write(out_pth + "/" + str(input_spp_list[i]) + "spp_nfor.exr", input_nfor)
def get_all_exr_dataset_one(dataset_dirs, suffix=""):
all_data = []
for dataset_dir in dataset_dirs:
files = os.listdir(dataset_dir)
files = [file for file in files if file.endswith(".exr")]
# files = [fn for fn in glob.glob(os.path.join(dataset_dir, suffix))]
filename = os.path.join(dataset_dir, files[0])
data = exr.read_all(filename)
all_data.append(data['default'][:, :, 0:3])
return np.array(all_data)
def save_all_exr_dataset(dataset_dirs, scene, target):
all_data = []
for dataset_dir in dataset_dirs:
files = os.listdir(dataset_dir)
files = [fn for fn in glob.glob(os.path.join(dataset_dir, '*.exr'))]
for f in files:
filename = os.path.join(dataset_dir, f)
data = exr.read_all(filename)
all_data.append(data['default'][:, :, 0:3])
exr.write(os.path.join('D:/training/', target, scene, f), data['default'][:, :, 0:3])
return np.array(all_data)
def load_normalize_one_exr_for_test(input_pth,
ref_pth,
BUFFER,
load_dtype="HALF",
color_merge=True):
channels = {
"diffuse": 3,
"specular": 3,
"albedo": 3,
"depth": 1,
"normal": 3,
"diffuseVariance": 1,
"specularVariance": 1,
"albedoVariance": 1,
"depthVariance": 1,
"normalVariance": 1
}
input_all_buffer = exr.read_all(input_pth, precision=load_dtype)
ref_all_buffer = exr.read_all(ref_pth, precision=load_dtype)
# test_depth = input_all_buffer['depth']
"저장 공간 설정 from sample"
total_ch = 0
ch_BUFFER = []
for b in BUFFER:
# h, w, ch = sample_dict[b].shape()
sample_data = input_all_buffer[b]
h, w, ch = sample_data.shape
total_ch += ch
ch_BUFFER.append(ch)
# color
if color_merge:
input_buffer = np.zeros((1, h, w, total_ch - 3),
dtype=sample_data.dtype)
ref_buffer = np.zeros((1, h, w, 3), dtype=sample_data.dtype)
input_buffer[0, :, :, :3] = norm.normalization_signed_log(
input_all_buffer["diffuse"] + input_all_buffer["specular"])
ref_buffer[0, :, :, :3] = norm.normalization_signed_log(
ref_all_buffer["diffuse"] + ref_all_buffer["specular"])
start_ch = 3
else:
input_buffer = np.zeros((1, h, w, total_ch), dtype=sample_data.dtype)
ref_buffer = np.zeros((1, h, w, 6), dtype=sample_data.dtype)
input_buffer[0, :, :, :3] = norm.normalization_signed_log(
input_all_buffer["diffuse"])
input_buffer[0, :, :, 3:6] = norm.normalization_signed_log(
input_all_buffer["specular"])
ref_buffer[0, :, :, :3] = norm.normalization_signed_log(
ref_all_buffer["diffuse"])
ref_buffer[0, :, :, 3:6] = norm.normalization_signed_log(
ref_all_buffer["specular"])
start_ch = 6
# g-buffer
for b in range(len(BUFFER)):
if BUFFER[b] == 'depth':
input_buffer[:, :, :, start_ch:start_ch +
channels[BUFFER[b]]] = norm.normalize_depth_1ch_v1(
input_all_buffer[BUFFER[b]])
start_ch += channels[BUFFER[b]]
elif BUFFER[b] == 'normal':
input_buffer[:, :, :, start_ch:start_ch +
channels[BUFFER[b]]] = norm.normalize_normal(
input_all_buffer[BUFFER[b]])
start_ch += channels[BUFFER[b]]
elif BUFFER[b] == 'albedo':
input_buffer[:, :, :, start_ch:start_ch +
channels[BUFFER[b]]] = input_all_buffer[BUFFER[b]]
start_ch += channels[BUFFER[b]]
return input_buffer, ref_buffer
def load_exrs_from_tungsten(DIR,
SCENE,
BUFFER,
mini_batch=True,
flag_saving=False,
saving_pth="tmp",
input_endswith="00128spp.exr",
ref_endswith="08192spp.exr",
load_dtype="HALF"):
"""
input : Dir = scene 폴더들이 있는 상위 폴더 위치, SCENE = 가져올 scene, BUFFER = 특정 feature 버퍼 이름
output : input buffer, ref buffer
특징 1 : original tungsten DB에서 원하는 input, ref 버퍼 만들기
특징 2 : 하나하나 읽어와 저장하는 구조라서 시간이 많이 걸림.
특징 3 : input = 124 spp, target = 8k spp 고정
특징 4 : 데이터의 양이 너무 많아서 굉장히 불러오는데 오래걸림.
"""
# DIR = 'D:/Tunsten_deep_learning_denoising_dataset/deep_learning_denoising/renderings'
# SCENE = ['bathroom2', 'car2', 'classroom', 'house', 'room2', 'room3', 'spaceship', 'staircase']
# BUFFER = ['diffuse', 'specular', 'albedo', 'depth', 'normal']
"path 불러오기"
ALL_SCENES = [] # 모든 spp, feature에 해당하는 buffer 이름
input_all_features = [] # [scene, files]
ref_all_feature = []
total_num_imgs = 0
for i in range(len(SCENE)):
# ALL_SCENES.append(os.path.join(DIR, SCENE[i]))
files = os.listdir(os.path.join(DIR, SCENE[i]))
input_all_features_one_scene = [
file for file in files if file.endswith(input_endswith)
]
ref_all_features_one_scene = [
file for file in files if file.endswith(ref_endswith)
]
if not mini_batch:
input_all_features.append(input_all_features_one_scene)
ref_all_feature.append(ref_all_features_one_scene)
total_num_imgs += len(input_all_features_one_scene)
else:
input_all_features.append(input_all_features_one_scene[:2])
ref_all_feature.append(ref_all_features_one_scene[:2])
total_num_imgs += 2
"저장 공간 설정 from sample"
sample_dict = exr.read_all(os.path.join(DIR, SCENE[0],
input_all_features[0][0]),
precision=load_dtype)
total_ch = 0
ch_BUFFER = []
for b in BUFFER:
# h, w, ch = sample_dict[b].shape()
sample_data = sample_dict[b]
h, w, ch = sample_data.shape
total_ch += ch
ch_BUFFER.append(ch)
input_buffer = np.ones((total_num_imgs, h, w, total_ch),
dtype=sample_data.dtype)
ref_buffer = np.ones((total_num_imgs, h, w, 6), dtype=sample_data.dtype)
"exr load"
img_index = 0
for s in range(len(input_all_features)):
for f in range(len(input_all_features[s])):
print(f)
"input : network input"
input_file_name = input_all_features[s][f]
one_input = exr.read_all(os.path.join(DIR, SCENE[s],
input_file_name),
precision=load_dtype)
start_ch = 0
for b in range(len(BUFFER)):
input_buffer[img_index, :, :, start_ch:start_ch +
ch_BUFFER[b]] = one_input[BUFFER[b]]
start_ch += ch_BUFFER[b]
"ref : 오직 color 부분만"
ref_file_name = ref_all_feature[s][f]
one_ref = exr.read_all(os.path.join(DIR, SCENE[s], ref_file_name),
precision=load_dtype)
ref_buffer[img_index, :, :, :3] = one_ref['diffuse']
ref_buffer[img_index, :, :, 3:] = one_ref['specular']
img_index += 1
if flag_saving:
# input_saving_name = "input_buffer"
# ref_saving_name = "ref_buffer"
#
# extension_name = ".npy"
# np.save(saving_pth + input_saving_name + extension_name, input_buffer)
# np.save(saving_pth + ref_saving_name + extension_name, ref_buffer)
start_ch = 0
extension_name = ".npy"
for b in range(len(BUFFER)):
input_saving_name = "input_" + BUFFER[b]
np.save(saving_pth + input_saving_name + extension_name,
input_buffer[:, :, :, start_ch:start_ch + ch_BUFFER[b]])
start_ch += ch_BUFFER[b]
np.save(saving_pth + "ref_diffuse" + extension_name,
ref_buffer[:, :, :, :3])
np.save(saving_pth + "ref_specular" + extension_name,
ref_buffer[:, :, :, 3:6])
# extension_name = ".h5"
# hf = h5py.File(saving_pth + ref_saving_name + extension_name, 'w')
# hf.create_dataset('input', data=input_buffer)
# hf.create_dataset('ref', data=ref_buffer)
# hf.close()
return input_buffer, ref_buffer