github/DeepFaceLab
1
0
Fork 0
mirror of https://github.com/iperov/DeepFaceLab.git synced 2025-08-22 06:23:20 -07:00
Code Issues Projects Releases Packages Wiki Activity Actions

fix: more testing

Browse source
This commit is contained in:
jh 2021-03-12 11:28:01 -08:00
commit 28279971a1
1 changed files with 16 additions and 21 deletions
Download patch file Download diff file Expand all files Collapse all files

37
core/leras/ops/__init__.py
View file

@ -308,7 +308,7 @@ def dssim(img1,img2, max_val, filter_size=11, filter_sigma=1.5, k1=0.01, k2=0.03
nn.dssim = dssim nn.dssim = dssim
def ms_ssim(resolution, kernel_size=11, k1=0.01, k2=0.03, max_value=1.0, def ms_ssim(img1, img2, resolution, kernel_size=11, k1=0.01, k2=0.03, max_value=1.0,
power_factors=(0.0448, 0.2856, 0.3001, 0.2363, 0.1333)): power_factors=(0.0448, 0.2856, 0.3001, 0.2363, 0.1333)):
# restrict mssim factors to those greater/equal to kernel size # restrict mssim factors to those greater/equal to kernel size
@ -318,29 +318,24 @@ def ms_ssim(resolution, kernel_size=11, k1=0.01, k2=0.03, max_value=1.0,
if sum(power_factors) < 1.0: if sum(power_factors) < 1.0:
power_factors = [x/sum(power_factors) for x in power_factors] power_factors = [x/sum(power_factors) for x in power_factors]
def loss(img1, img2): img_dtype = img1.dtype
img_dtype = img1.dtype if img_dtype != img2.dtype:
if img_dtype != img2.dtype: raise ValueError("img1.dtype != img2.dtype")
raise ValueError("img1.dtype != img2.dtype")
if img_dtype != tf.float32: if img_dtype != tf.float32:
img1 = tf.cast(img1, tf.float32) img1 = tf.cast(img1, tf.float32)
img2 = tf.cast(img2, tf.float32) img2 = tf.cast(img2, tf.float32)
# Transpose images from NCHW to NHWC
img1_t = tf.transpose(img1, [0, 2, 3, 1])
img2_t = tf.transpose(img2, [0, 2, 3, 1])
ms_ssim_val = tf.image.ssim_multiscale(img1_t, img2_t, max_val=max_value, power_factors=power_factors,
filter_size=kernel_size, k1=k1, k2=k2)
ms_ssim_loss = (1.0 - ms_ssim_val) / 2.0
if img_dtype != tf.float32:
# Transpose images from NCHW to NHWC ms_ssim_loss = tf.cast(ms_ssim_loss, img_dtype)
img1_t = tf.transpose(img1, [0, 2, 3, 1]) return ms_ssim_loss
img2_t = tf.transpose(img2, [0, 2, 3, 1])
ms_ssim_val = tf.image.ssim_multiscale(img1_t, img2_t, max_val=max_value, power_factors=power_factors,
filter_size=kernel_size, k1=k1, k2=k2)
ms_ssim_loss = (1.0 - ms_ssim_val) / 2.0
if img_dtype != tf.float32:
ms_ssim_loss = tf.cast(ms_ssim_loss, img_dtype)
return ms_ssim_loss
return loss
nn.ms_ssim = ms_ssim nn.ms_ssim = ms_ssim