15.04.2025, 17:57
@Dan64:
Had a quick look at the new autoadjust.
I like the new versions. (just a bit slow)
a. Do they all have to be RGB24 only? iirc. cv2 should be able to also use RGB48
b. wouldn't it be better to not use PIL ?
c. have you tried cupy?
d. It seems wrong to use range_s="full" and later use tv range restrictions (16,235) for the limits
Here's a quick try for a high bit depth version without PIL:
I'm not sure about the luma range stuff. 
I think it would be better to: Require full range RGB input and only work with full range luma.
Just for inspiration (probably still got some errors in it):
Cu Selur
Had a quick look at the new autoadjust.
I like the new versions. (just a bit slow)
a. Do they all have to be RGB24 only? iirc. cv2 should be able to also use RGB48
b. wouldn't it be better to not use PIL ?
c. have you tried cupy?
d. It seems wrong to use range_s="full" and later use tv range restrictions (16,235) for the limits
Here's a quick try for a high bit depth version without PIL:
def AutoGainTest(clip: vs.VideoNode, clip_limit: float = 1.0, strength: float = 0.5) -> vs.VideoNode:
if not isinstance(clip, vs.VideoNode):
raise vs.Error("AutoGain: This is not a clip")
# Convert to RGB if needed (supporting high bit depth)
rgb_format = vs.RGBS if clip.format.bits_per_sample > 8 else vs.RGB24
if clip.format.id != rgb_format:
if clip.format.color_family == vs.YUV:
rgb_clip = clip.resize.Bicubic(format=rgb_format, matrix_in_s="709", range_s="full")
else:
rgb_clip = clip.resize.Bicubic(format=rgb_format, range_s="full")
else:
rgb_clip = clip
weight = max(min(1.0 - strength, 1.0), 0.0)
bits = clip.format.bits_per_sample
max_val = (1 << bits) - 1
loose_max_limit = (235 + 1) << (bits - 8) if bits <= 16 else (235 + 1) * 256
loose_min_limit = 16 << (bits - 8) if bits <= 16 else 16 * 256
clip_limit_factor = (1 - clip_limit/100.0) if clip_limit > 0 else 1.0
def frame_autogain(n, f):
# Create numpy array from frame planes (handles both 8-bit and high bit depth)
if rgb_format == vs.RGB24:
img_np = np.stack([
np.asarray(f[0]),
np.asarray(f[1]),
np.asarray(f[2])
], axis=2).astype(np.float32)
else:
# For high bit depth (RGBS)
img_np = np.stack([
np.asarray(f[0]) * 255,
np.asarray(f[1]) * 255,
np.asarray(f[2]) * 255
], axis=2).astype(np.float32)
# Process image
yuv = cv2.cvtColor(img_np, cv2.COLOR_RGB2YUV)
dY = yuv[:, :, 0]
maxY = min(dY.max(), loose_max_limit)
minY = max(dY.min(), loose_min_limit)
y_range = maxY - minY
if y_range > 0:
scale = (loose_max_limit - loose_min_limit) / y_range
y_offset = (loose_min_limit - scale * minY)
y_gain = (scale - 1.0)
# Apply gain and offset
yuv[:, :, 0] = np.clip((dY + y_offset) * (y_gain + 1), 0, 255)
# Convert back to RGB
rgb_new = cv2.cvtColor(yuv, cv2.COLOR_YUV2RGB)
# Create new frame
new_frame = f.copy()
if rgb_format == vs.RGB24:
for i in range(3):
np.copyto(np.asarray(new_frame[i]), rgb_new[:, :, i].astype(np.uint8))
else:
# For high bit depth (RGBS)
for i in range(3):
np.copyto(np.asarray(new_frame[i]), (rgb_new[:, :, i] / 255).astype(np.float32))
return new_frame
clip_a = rgb_clip.std.ModifyFrame(clips=[rgb_clip], selector=frame_autogain)
clip_rgb = core.std.Merge(clip_a, rgb_clip, weight) if weight > 0 else clip_a
# Convert back to original format if needed
if clip.format.id != rgb_format:
if clip.format.color_family == vs.YUV:
return clip_rgb.resize.Bicubic(format=clip.format.id, matrix_s="709", range_s="limited")
else:
return clip_rgb.resize.Bicubic(format=clip.format.id, range_s=clip.get_frame(0).props._ColorRange)
return clip_rgbdef AutoGainFull(clip: vs.VideoNode, clip_limit: float = 1.0, strength: float = 0.5) -> vs.VideoNode:
"""
AutoGain filter for full-range RGB input/output only.
Uses RGB limits (0-max) instead of YUV TV range (16-235).
Args:
clip: RGB input clip (must be full range)
clip_limit: Threshold for contrast limiting (0-50)
strength: Filter strength (0-1)
"""
if not isinstance(clip, vs.VideoNode):
raise vs.Error("AutoGain: Input must be a clip")
# Verify input is RGB and full range
if clip.format.color_family != vs.RGB:
raise vs.Error("AutoGain: Input must be RGB format")
if hasattr(clip.get_frame(0).props, "_ColorRange") and clip.get_frame(0).props._ColorRange != 0:
raise vs.Error("AutoGain: Input must be full range (0-255/1023/etc.)")
bits = clip.format.bits_per_sample
max_val = (1 << bits) - 1
scale_factor = 255 / max_val if bits > 8 else 1
# Parameters
weight = max(min(1.0 - strength, 1.0), 0.0)
clip_limit_factor = (1 - clip_limit/100.0) if clip_limit > 0 else 1.0
def frame_autogain(n, f):
# Create numpy array from frame planes (handles all bit depths)
if clip.format.sample_type == vs.INTEGER:
img_np = np.stack([
np.asarray(f[0]),
np.asarray(f[1]),
np.asarray(f[2])
], axis=2).astype(np.float32) * scale_factor
else:
# Floating point format (RGBS)
img_np = np.stack([
np.asarray(f[0]) * 255,
np.asarray(f[1]) * 255,
np.asarray(f[2]) * 255
], axis=2).astype(np.float32)
# Convert to YUV for processing (still using full range)
yuv = cv2.cvtColor(img_np, cv2.COLOR_RGB2YUV)
dY = yuv[:, :, 0]
# Calculate using full range (0-255)
maxY = dY.max()
minY = dY.min()
y_range = maxY - minY
if y_range > 0:
scale = 255 / y_range # Full range scaling
y_offset = -scale * minY
y_gain = (scale - 1.0)
# Apply with clip_limit factor
y_offset *= clip_limit_factor
y_gain *= clip_limit_factor
# Apply gain and offset
yuv[:, :, 0] = np.clip((dY + y_offset) * (y_gain + 1), 0, 255)
# Convert back to RGB
rgb_new = cv2.cvtColor(yuv, cv2.COLOR_YUV2RGB)
# Create new frame with proper bit depth
new_frame = f.copy()
if clip.format.sample_type == vs.INTEGER:
output = np.clip(rgb_new / scale_factor, 0, max_val).astype(np.uint16 if bits > 8 else np.uint8)
for i in range(3):
np.copyto(np.asarray(new_frame[i]), output[:, :, i])
else:
# Floating point format (RGBS)
for i in range(3):
np.copyto(np.asarray(new_frame[i]), (rgb_new / 255).astype(np.float32))
return new_frame
# Process and optionally blend with original
clip_a = clip.std.ModifyFrame(clips=[clip], selector=frame_autogain)
return core.std.Merge(clip_a, clip, weight) if weight > 0 else clip_aI think it would be better to: Require full range RGB input and only work with full range luma.
Just for inspiration (probably still got some errors in it):
def AutoGainYUV(clip: vs.VideoNode, clip_limit: float = 1.0, strength: float = 0.5, planes: tuple = (0,)) -> vs.VideoNode:
"""
YUV-only AutoGain that processes luma (and optionally chroma) directly
Requires YUV input, avoids all RGB conversions
Args:
clip: YUV input clip
clip_limit: Threshold for contrast limiting (0-50)
strength: Filter strength (0-1)
planes: Which planes to process (default=(0,) for luma only)
"""
if not isinstance(clip, vs.VideoNode):
raise vs.Error("AutoGainYUV: Input must be a clip")
if clip.format.color_family != vs.YUV:
raise vs.Error("AutoGainYUV: Input must be YUV format")
bits = clip.format.bits_per_sample
is_float = clip.format.sample_type == vs.FLOAT
weight = max(min(1.0 - strength, 1.0), 0.0)
clip_limit_factor = (1 - clip_limit/100.0) if clip_limit > 0 else 1.0
def frame_autogain(n, f):
# Get frame properties to determine data type
sample_type = clip.format.sample_type
bytes_per_sample = clip.format.bytes_per_sample
# Process each plane
processed_planes = []
for i in range(clip.format.num_planes):
if i not in planes:
processed_planes.append(np.asarray(f[i]))
continue
# Convert plane to numpy array with correct dtype
if sample_type == vs.FLOAT:
plane = np.asarray(f[i], dtype=np.float32)
processing_plane = plane * 255
elif bytes_per_sample == 1:
plane = np.asarray(f[i], dtype=np.uint8)
processing_plane = plane.astype(np.float32)
else:
plane = np.asarray(f[i], dtype=np.uint16)
processing_plane = plane.astype(np.float32)
# Calculate statistics
maxY = processing_plane.max()
minY = processing_plane.min()
y_range = maxY - minY
if y_range > 0:
scale = 255 / y_range
y_offset = -scale * minY
y_gain = (scale - 1.0)
# Apply with clip_limit factor
y_offset *= clip_limit_factor
y_gain *= clip_limit_factor
# Apply gain and offset
processing_plane = np.clip((processing_plane + y_offset) * (y_gain + 1), 0, 255)
# Convert back to original format
if sample_type == vs.FLOAT:
result = processing_plane / 255
elif bytes_per_sample == 1:
result = np.clip(processing_plane, 0, 255).astype(np.uint8)
else:
result = np.clip(processing_plane, 0, 65535).astype(np.uint16)
processed_planes.append(result)
# Create new frame with processed planes
new_frame = f.copy()
for i in range(clip.format.num_planes):
np.copyto(np.asarray(new_frame[i]), processed_planes[i])
return new_frame
# Process and optionally blend with original
clip_a = clip.std.ModifyFrame(clips=[clip], selector=frame_autogain)
return core.std.Merge(clip_a, clip, weight) if weight > 0 else clip_aCu Selur
----
Dev versions are in the 'experimental'-folder of my GoogleDrive, which is linked on the download page.
Dev versions are in the 'experimental'-folder of my GoogleDrive, which is linked on the download page.