Public/automatic
1
0
Fork 0
mirror of https://github.com/vladmandic/automatic synced 2026-09-19 01:04:32 +02:00
Code Issues Packages Projects Releases Wiki Activity

update hdr

Browse Source
This commit is contained in:
Vladimir Mandic
2023-11-27 09:03:22 -05:00
parent 49f8730814
commit 12411f2858
7 changed files with 52 additions and 34 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/processing_correction.py
+14 -8
View File
@@ -11,11 +11,12 @@ from modules import shared
debug = shared.log.info if os.environ.get('SD_HDR_DEBUG', None) is not None else lambda *args, **kwargs: None
def soft_clamp_tensor(input_tensor, threshold=3.5, boundary=4):
def soft_clamp_tensor(input_tensor, threshold=0.8, boundary=4):
# shrinking towards the mean; will also remove outliers
if max(abs(input_tensor.max()), abs(input_tensor.min())) < 4:
if max(abs(input_tensor.max()), abs(input_tensor.min())) < boundary or threshold == 0:
return input_tensor
channel_dim = 1
threshold *= boundary
max_vals = input_tensor.max(channel_dim, keepdim=True)[0]
max_replace = ((input_tensor - threshold) / (max_vals - threshold)) * (boundary - threshold) + threshold
over_mask = input_tensor > threshold
@@ -23,21 +24,26 @@ def soft_clamp_tensor(input_tensor, threshold=3.5, boundary=4):
min_replace = ((input_tensor + threshold) / (min_vals + threshold)) * (-boundary + threshold) - threshold
under_mask = input_tensor < -threshold
debug(f'HDE soft clamp: threshold={threshold} boundary={boundary}')
res = torch.where(over_mask, max_replace, torch.where(under_mask, min_replace, input_tensor))
return res
input_tensor = torch.where(over_mask, max_replace, torch.where(under_mask, min_replace, input_tensor))
return input_tensor
def center_tensor(input_tensor, channel_shift=1.0, full_shift=1.0, channels=[0, 1, 2, 3]):
if channel_shift == 0 and full_shift == 0:
return input_tensor
means = []
for channel in channels:
means.append(input_tensor[0, channel].mean())
input_tensor[0, channel] -= means[-1] * channel_shift
debug(f'HDR center: channel-shift{channel_shift} full-shift={full_shift} means={torch.stack(means)}')
res = input_tensor - input_tensor.mean() * full_shift
return res
input_tensor = input_tensor - input_tensor.mean() * full_shift
return input_tensor
def maximize_tensor(input_tensor, boundary=4.0, channels=[0, 1, 2]):
def maximize_tensor(input_tensor, boundary=1.0, channels=[0, 1, 2]):
if boundary == 1.0:
return input_tensor
boundary *= 4
min_val = input_tensor.min()
max_val = input_tensor.max()
normalization_factor = boundary / max(abs(min_val), abs(max_val))
@@ -52,6 +58,6 @@ def correction_callback(p, timestep, kwags):
if timestep > 700 and p.hdr_center:
kwags["latents"] = center_tensor(kwags["latents"], channel_shift=p.hdr_channel_shift, full_shift=p.hdr_full_shift)
if timestep > 1 and timestep < 100 and p.hdr_maximize:
kwags["latents"] = center_tensor(kwags["latents"], channel_shift=0.6, full_shift=1.0)
kwags["latents"] = center_tensor(kwags["latents"], channel_shift=p.hdr_max_center, full_shift=1.0)
kwags["latents"] = maximize_tensor(kwags["latents"], boundary=p.hdr_max_boundry)
return kwags