def generateLUT1DInverseIndexMap(resolution, samples, minInputValue,
maxInputValue):
lutpns = []
# Invert happens in 3 stages
# 1. Create the index map that goes from LUT output values to index values
# 2. Create a LUT that maps from index values to [0,1]
# 3. Create a Range node to remap from [0,1] to [minInput,maxInput]
# Get the resolution of the prelut
inputResolution = resolution[0]
channels = resolution[1]
#print( inputResolution, minInputValue, maxInputValue )
# Index Maps for the inverse LUT
indexMaps = []
for c in range(channels):
indexMapInput = [0.0] * inputResolution
for i in range(inputResolution):
indexMapInput[i] = samples[i * channels + c]
indexMapOutput = range(inputResolution)
indexMaps.append([indexMapInput, indexMapOutput])
# Sample values for the LUT - output is [0,1]
inverseSamples = [0.0] * inputResolution * channels
for i in range(inputResolution):
v = float(i) / (inputResolution - 1)
for c in range(channels):
inverseSamples[i * channels + c] = v
# Create a 1D LUT with generated index map and sample values
lutpn = clf.LUT1D(clf.bitDepths["FLOAT16"], clf.bitDepths["FLOAT16"],
"inverse_1d_lut", "inverse_1d_lut")
if channels == 3:
lutpn.setIndexMaps(indexMaps[0], indexMaps[1], indexMaps[2])
else:
lutpn.setIndexMaps(indexMaps[0])
lutpn.setArray(channels, inverseSamples)
lutpns.append(lutpn)
# Create a Range node to expaand from [0,1] to [minIn, maxIn]
if minInputValue != 0.0 or maxInputValue != 1.0:
rangepn2 = clf.Range(clf.bitDepths["FLOAT16"],
clf.bitDepths["FLOAT16"], "inverse_1d_range_1",
"inverse_1d_range_1")
rangepn2.setMinInValue(0.0)
rangepn2.setMaxInValue(1.0)
rangepn2.setMinOutValue(minInputValue)
rangepn2.setMaxOutValue(maxInputValue)
lutpns.append(rangepn2)
return lutpns
def generateLUT1DInverseResampled(resolution, samples, minInputValue,
maxInputValue):
lutpns = []
# Invert happens in 3 stages
# 1. Range values down from the min and max output values to 0-1
# 2. Generate the inverse LUT for these newly reranged values
# 3. Range the value up to the range defined by minInputValue and maxInputValue
# This is similar to how .CSP preluts are turned into ProcessNodes
# Get the resolution of the prelut
inputResolution = resolution[0]
channels = resolution[1]
# XXX
# Given that we're resampling, we should probably increase the
# resolution of the resampled lut relative to the source
outputResolution = inputResolution
outputResolution *= 2
# Find the minimum and maximum input
# XXX
# We take the min and max of all three preluts because the Range node
# only takes single value, not RGB triples. If the prelut ranges are
# very different, this could introduce some artifacting
minOutputValue = samples[0]
for c in range(channels):
minOutputValue = min(minOutputValue, samples[c])
maxOutputValue = samples[-channels]
for c in range(channels):
maxOutputValue = max(maxOutputValue, samples[-channels + c])
#print( inputResolution, minInputValue, maxInputValue, minOutputValue, maxOutputValue )
# Create a Range node to normalize data from the range [minOut, maxOut]
rangepn1 = clf.Range(clf.bitDepths["FLOAT16"], clf.bitDepths["FLOAT16"],
"inverse_1d_range_1", "inverse_1d_range_1")
rangepn1.setMinInValue(minOutputValue)
rangepn1.setMaxInValue(maxOutputValue)
rangepn1.setMinOutValue(0.0)
rangepn1.setMaxOutValue(1.0)
lutpns.append(rangepn1)
# Generate inverse 1d LUT by running values through the
# - inverse normalization [0,1] back to [minOut, maxOut]
# - the inverse of the original LUT
inverseSamples = [0.0] * outputResolution * channels
for inverseLutIndex in range(outputResolution):
# Normalized LUT input
inputValue = float(inverseLutIndex) / (outputResolution - 1)
# Invert the normalization
rangedValue = inputValue * (maxOutputValue -
minOutputValue) + minOutputValue
inverseSample = [0.0] * channels
# For each channel
for channel in range(channels):
# Find the location of the de-normalized value in the lut
for lutIndex in range(inputResolution):
sampleIndex = lutIndex * channels + channel
if samples[sampleIndex] > rangedValue:
break
# Get the interpolation value
lutIndexLow = max(0, lutIndex - 1)
lutIndexHigh = min(inputResolution - 1, lutIndex)
sampleIndexLow = lutIndexLow * channels + channel
sampleIndexHigh = lutIndexHigh * channels + channel
if lutIndexLow == lutIndexHigh:
lutInterp = 0.0
else:
lutInterp = (rangedValue - samples[sampleIndexLow]) / (
samples[sampleIndexHigh] - samples[sampleIndexLow])
# Find the output value
outputInterpolated = (lutInterp + lutIndexLow) / (inputResolution -
1)
inverseSample[channel] = outputInterpolated
inverseSamples[inverseLutIndex * channels:(inverseLutIndex + 1) *
channels] = inverseSample
# Create a 1D LUT with generated sample values
lutpn = clf.LUT1D(clf.bitDepths["FLOAT16"], clf.bitDepths["FLOAT16"],
"inverse_1d_lut", "inverse_1d_lut")
lutpn.setArray(channels, inverseSamples)
lutpns.append(lutpn)
# Create a Range node to expaand from [0,1] to [minIn, maxIn]
rangepn2 = clf.Range(clf.bitDepths["FLOAT16"], clf.bitDepths["FLOAT16"],
"inverse_1d_range_2", "inverse_1d_range_2")
rangepn2.setMinInValue(0.0)
rangepn2.setMaxInValue(1.0)
rangepn2.setMinOutValue(minInputValue)
rangepn2.setMaxOutValue(maxInputValue)
lutpns.append(rangepn2)
return lutpns
def generateLUT1DInverseHalfDomain(resolution,
samples,
minInputValue,
maxInputValue,
rawHalfs=False):
lutpns = []
# Invert happens in 1 stages
# 1. Generate the inverse LUT for each possible half-float value
# Get the resolution of the prelut
inputResolution = resolution[0]
channels = resolution[1]
# For this inversion approach, we will always use 64k entries
outputResolution = 65536
#print( inputResolution, minInputValue, maxInputValue, minOutputValue, maxOutputValue )
# Generate inverse 1d LUT by running values through the
# - the inverse of the original LUT
inverseSamples = [0.0] * outputResolution * channels
for inverseLutIndex in range(outputResolution):
# LUT input
inputValue = uint16ToHalf(inverseLutIndex)
inverseSample = [0.0] * channels
# For each channel
for channel in range(channels):
if math.isnan(inputValue):
outputInterpolated = inputValue
elif math.isinf(inputValue):
outputInterpolated = inputValue
else:
# Find the location of the value in the lut
for lutIndex in range(inputResolution):
sampleIndex = lutIndex * channels + channel
if samples[sampleIndex] > inputValue:
break
# Get the interpolation value
lutIndexLow = max(0, lutIndex - 1)
lutIndexHigh = min(inputResolution - 1, lutIndex)
sampleIndexLow = lutIndexLow * channels + channel
sampleIndexHigh = lutIndexHigh * channels + channel
if lutIndexLow == lutIndexHigh:
lutInterp = 0.0
else:
lutInterp = (inputValue - samples[sampleIndexLow]) / (
samples[sampleIndexHigh] - samples[sampleIndexLow])
# Find the output value
outputInterpolated = (lutInterp +
lutIndexLow) / (inputResolution - 1)
inverseSample[channel] = outputInterpolated
inverseSamples[inverseLutIndex * channels:(inverseLutIndex + 1) *
channels] = inverseSample
# Create a 1D LUT with generated sample values
if rawHalfs:
lutpn = clf.LUT1D(clf.bitDepths["FLOAT16"],
clf.bitDepths["FLOAT16"],
"inverse_1d_lut",
"inverse_1d_lut",
rawHalfs=rawHalfs,
halfDomain=True)
else:
lutpn = clf.LUT1D(clf.bitDepths["FLOAT16"],
clf.bitDepths["FLOAT16"],
"inverse_1d_lut",
"inverse_1d_lut",
halfDomain=True)
lutpn.setArray(channels, inverseSamples)
lutpns.append(lutpn)
return lutpns
def write_CLF_1d_3d_1d(lutPath,
samples1dIn,
lutResolution1dIn,
inputMin,
inputMax,
samples3d,
lutResolution3d,
samples1dOut,
lutResolution1dOut,
outputMin,
outputMax,
inversesUseIndexMaps=True,
inversesUseHalfDomain=True):
lutpns = []
# Create the input shaper
if samples1dIn:
if inversesUseIndexMaps:
print("Generating inverse of 1D LUT using Index Maps")
lutpnInverses = convertLUTtoCLF.generateLUT1DInverseIndexMap(
[lutResolution1dIn, 3], samples1dIn, inputMin, inputMax)
elif inversesUseHalfDomain:
print("Generating full half-domain inverse of 1D LUT")
lutpnInverses = convertLUTtoCLF.generateLUT1DInverseHalfDomain(
[lutResolution1dIn, 3],
samples1dIn,
inputMin,
inputMax,
rawHalfs=True)
else:
print("Generating resampled inverse of 1D LUT")
lutpnInverses = convertLUTtoCLF.generateLUT1DInverseResampled(
[lutResolution1dIn, 3], samples1dIn, inputMin, inputMax)
lutpns.extend(lutpnInverses)
# Create the 3D LUT
clfSamples = [0.0] * (lutResolution3d[0] * lutResolution3d[1] *
lutResolution3d[2]) * 3
index = 0
for r in range(lutResolution3d[0]):
for g in range(lutResolution3d[1]):
for b in range(lutResolution3d[2]):
for c in range(3):
clfSamples[index] = samples3d[r][g][b][c]
index += 1
interpolation = 'trilinear'
lut3dpn = clf.LUT3D(clf.bitDepths["FLOAT16"],
clf.bitDepths["FLOAT16"],
"lut3d",
"lut3d",
interpolation=interpolation)
lut3dpn.setArray(
[lutResolution3d[0], lutResolution3d[1], lutResolution3d[2]],
clfSamples)
lutpns.append(lut3dpn)
# Create the output shaper
if samples1dOut:
interpolation = 'linear'
lutpn = clf.LUT1D(clf.bitDepths["FLOAT16"],
clf.bitDepths["FLOAT16"],
"lut1d",
"lut1d",
interpolation=interpolation)
lutpn.setArray(3, samples1dOut)
lutpns.append(lutpn)
# Wrap in a ProcessList and write to disk
pl = clf.ProcessList()
# Populate
pl.setID('Converted lut')
pl.setCompCLFversion(1.0)
pl.setName('Converted lut')
for lutpn in lutpns:
pl.addProcess(lutpn)
# Write CLF to disk
pl.writeFile(lutPath)
def readCSP(lutPath, direction='forward', interpolation='linear'):
with open(lutPath) as f:
lines = f.read().splitlines()
dataFormat = LUTFORMAT_UNKNOWN
resolution = []
samples = []
prelut = []
minInputValue = 0.0
maxInputValue = 1.0
tokens = lines[0].split()
if tokens[0] == "CSPLUTV100":
format = lines[1].split()[0]
if format in ["1D", "3D"]:
# Find the first line of metadata
metaStart = 2
while lines[metaStart].rstrip() != "BEGIN METADATA":
metaStart += 1
metaStart += 1
metadata = ""
dataStart = metaStart
for line in lines[metaStart:]:
dataStart += 1
if line.rstrip() == "END METADATA":
break
else:
metadata += (line.rstrip())
#print( "metadata line : %s" % line.rstrip() )
#print( "metadata : %s" % metadata)
dataStart += 1
while lines[dataStart].rstrip() == '':
#print( "blank line")
dataStart += 1
#print( "Prelut data starts on line : %d" % dataStart )
#
# Read Index Maps
#
# Red Index Map
prelutRedResolution = int(lines[dataStart + 0])
prelutRedInput = map(float, lines[dataStart + 1].split())
prelutRedOutput = map(float, lines[dataStart + 2].split())
# Green Index Map
prelutGreenResolution = int(lines[dataStart + 3])
prelutGreenInput = map(float, lines[dataStart + 4].split())
prelutGreenOutput = map(float, lines[dataStart + 5].split())
# Blue Index Map
prelutBlueResolution = int(lines[dataStart + 6])
prelutBlueInput = map(float, lines[dataStart + 7].split())
prelutBlueOutput = map(float, lines[dataStart + 8].split())
prelut = [[prelutRedInput, prelutRedOutput],
[prelutGreenInput, prelutGreenOutput],
[prelutBlueInput, prelutBlueOutput]]
#
# Read LUT data
#
dataStart = dataStart + 10
while lines[dataStart].rstrip() == '':
#print( "blank line")
dataStart += 1
resolution = map(int, lines[dataStart].split())
dataStart += 1
#print( "lut data starts on line : %d" % dataStart )
# 3D LUT data
if format == "3D":
dataFormat = LUTFORMAT_3D
# CSP incremements LUT samples red, then green, then blue
# CLF incremements LUT samples blue, then green, then red
# so we need to move samples around
samples = [
0.0
] * resolution[0] * resolution[1] * resolution[2] * 3
cspIndex = 0
for line in lines[dataStart:]:
# Convert from sample number to LUT index, CSP-style
indexR = cspIndex % resolution[0]
indexG = (cspIndex / resolution[0]) % resolution[1]
indexB = cspIndex / (resolution[0] * resolution[1])
# Convert from LUT index to sample number, CLF-style
clfIndex = (indexR * resolution[0] +
indexG) * resolution[1] + indexB
clfIndex *= 3
# Convert from text to float
cspSamples = map(float, line.split())
# Add to the sample values array
if len(cspSamples) == 3:
#print( "csp sample %d -> lut index : %d, %d, %d -> clf index : %d" %
# (cspIndex, indexR, indexG, indexB, clfIndex))
#print( "lut value : %3.6f %3.6f %3.6f" % (cspSamples[0], cspSamples[1], cspSamples[2]) )
samples[clfIndex:clfIndex + 3] = cspSamples
cspIndex += 1
# 1D LUT data
elif format == "1D":
dataFormat = LUTFORMAT_1D
samples = [0.0] * resolution[0] * 3
cspIndex = 0
for line in lines[dataStart:]:
clfIndex = cspIndex
clfIndex *= 3
# Convert from text to float
cspSamples = map(float, line.split())
# Add to the sample values array
if len(cspSamples) == 3:
#print( "csp sample %d -> clf index : %d" %
# (cspIndex, clfIndex/3))
#print( "lut value : %3.6f %3.6f %3.6f" % (cspSamples[0], cspSamples[1], cspSamples[2]) )
samples[clfIndex:clfIndex + 3] = cspSamples
cspIndex += 1
#
# Create ProcessNodes
#
lutpns = []
if prelut != []:
print("Generating prelut")
prelutpns = generateCLFPrelut(prelut)
if prelutpns != []:
for prelutpn in prelutpns:
lutpns.append(prelutpn)
if format == "3D":
lutpn = clf.LUT3D(clf.bitDepths["FLOAT16"], clf.bitDepths["FLOAT16"],
"lut3d", "lut3d")
lutpn.setArray(resolution, samples)
lutpns.append(lutpn)
elif format == "1D":
lutpn = clf.LUT1D(clf.bitDepths["FLOAT16"], clf.bitDepths["FLOAT16"],
"lut1d", "lut1d")
lutpn.setArray(3, samples)
lutpns.append(lutpn)
#print (dataFormat, resolution, samples, prelut, minInputValue, maxInputValue)
return lutpns
def generateCLFPrelut(cspPreluts):
prelutpns = []
# Get the individual preluts
(prelutR, prelutG, prelutB) = cspPreluts
# Get the resolution of the prelut
inputResolution = max(len(prelutR[0]), len(prelutG[0]), len(prelutB[0]))
# XXX
# Given that we're resampling, we should probably increase the
# resolution of the resampled lut relative to the source
outputResolution = inputResolution
# If the prelut only affects the range, skip this step
if inputResolution > 2:
outputResolution *= 2
# Find the minimum and maximum input
# XXX
# We take the min and max of all three preluts because the Range node
# only takes single value, not RGB triples. If the prelut ranges are
# very different, this could introduce some artifacting
minInputValue = min(prelutR[0][0], prelutG[0][0], prelutB[0][0])
maxInputValue = max(prelutR[0][-1], prelutG[0][-1], prelutB[0][-1])
#print( inputResolution, minInputValue, maxInputValue )
# Create a Range node to normalize data from that range [min, max]
rangepn = clf.Range(clf.bitDepths["FLOAT16"], clf.bitDepths["FLOAT16"],
"prelut_range", "prelut_range")
rangepn.setMinInValue(minInputValue)
rangepn.setMaxInValue(maxInputValue)
rangepn.setMinOutValue(0.0)
rangepn.setMaxOutValue(1.0)
prelutpns.append(rangepn)
# If the prelut only affects the range, skip generating a lut to represent it
if inputResolution > 2:
# Generate 1d LUT by running values through the
# - inverse normalization
# - the cspprelut
samples = [0.0] * outputResolution * 3
for i in range(outputResolution):
# Normalized LUT input
inputValue = float(i) / (outputResolution - 1)
# Invert the normalization
rangedValue = inputValue * (maxInputValue -
minInputValue) + minInputValue
sample = [0.0, 0.0, 0.0]
# For each channel
for channel in range(len(cspPreluts)):
# Find the location of the de-normalized value in the prelut
for prelutIndex in range(inputResolution):
if cspPreluts[channel][0][prelutIndex] > rangedValue:
break
# Get the interpolation value
prelutIndexLow = max(0, prelutIndex - 1)
prelutIndexHigh = min(inputResolution - 1, prelutIndex)
prelutInterp = (rangedValue -
cspPreluts[channel][0][prelutIndexLow]) / (
cspPreluts[channel][0][prelutIndexHigh] -
cspPreluts[channel][0][prelutIndexLow])
# Find the output value
outputInterpolationRange = (
cspPreluts[channel][1][prelutIndexHigh] -
cspPreluts[channel][1][prelutIndexLow])
outputInterpolated = prelutInterp * outputInterpolationRange + cspPreluts[
channel][1][prelutIndexLow]
sample[channel] = outputInterpolated
samples[i * 3:(i + 1) * 3] = sample
# Create a 1D LUT with generated sample values
lutpn = clf.LUT1D(clf.bitDepths["FLOAT16"], clf.bitDepths["FLOAT16"],
"prelut_lut1d", "prelut_lut1d")
lutpn.setArray(len(cspPreluts), samples)
prelutpns.append(lutpn)
return prelutpns
def readSPI1D(lutPath,
direction='forward',
interpolation='linear',
inversesUseIndexMaps=True,
inversesUseHalfDomain=True):
with open(lutPath) as f:
lines = f.read().splitlines()
#
# Read LUT data
#
dataFormat = LUTFORMAT_1D
resolution = [0, 0]
samples = []
indexMap = []
minInputValue = 0.0
maxInputValue = 1.0
for line in lines:
#print( "line : %s" % line )
tokens = line.split()
if tokens[0] == "Version":
version = int(tokens[1])
if version != 1:
break
elif tokens[0] == "From":
minInputValue = float(tokens[1])
maxInputValue = float(tokens[2])
elif tokens[0] == "Length":
resolution[0] = int(tokens[1])
elif tokens[0] == "Components":
resolution[1] = int(tokens[1])
elif tokens[0] in ["{", "}"]:
continue
else:
samples.extend(map(float, tokens))
#else:
# print( "Skipping line : %s" % tokens )
#
# Create ProcessNodes
#
lutpns = []
# Forward transform, pretty straightforward
if direction == 'forward':
# Remap input range
if minInputValue != 0.0 or maxInputValue != 1.0:
rangepn = clf.Range(clf.bitDepths["FLOAT16"],
clf.bitDepths["FLOAT16"], "range", "range")
rangepn.setMinInValue(minInputValue)
rangepn.setMaxInValue(maxInputValue)
rangepn.setMinOutValue(0.0)
rangepn.setMaxOutValue(1.0)
lutpns.append(rangepn)
# LUT node
lutpn = clf.LUT1D(clf.bitDepths["FLOAT16"],
clf.bitDepths["FLOAT16"],
"lut1d",
"lut1d",
interpolation=interpolation)
lutpn.setArray(resolution[1], samples)
lutpns.append(lutpn)
# Inverse transform, LUT has to be resampled
else:
if inversesUseIndexMaps:
print("Generating inverse of 1D LUT using Index Maps")
lutpnInverses = generateLUT1DInverseIndexMap(
resolution, samples, minInputValue, maxInputValue)
elif inversesUseHalfDomain:
print("Generating full half-domain inverse of 1D LUT")
lutpnInverses = generateLUT1DInverseHalfDomain(resolution,
samples,
minInputValue,
maxInputValue,
rawHalfs=True)
else:
print("Generating resampled inverse of 1D LUT")
lutpnInverses = generateLUT1DInverseResampled(
resolution, samples, minInputValue, maxInputValue)
lutpns.extend(lutpnInverses)
#print (dataFormat, resolution, samples, indexMap, minInputValue, maxInputValue)
return lutpns