mirror of
https://github.com/alicevision/Meshroom.git
synced 2025-04-30 02:37:26 +02:00
e76dbd2a4f
Outliers are now supported for the HDR fusion nodes and are excluded from the computations as soon as they have been detected. However, the chunks' size computation still includes them, as it simply uses the number of detected brackets and the total number of input images. With this commit, the detected outliers are excluded from the chunks' size computation, thus preventing any error that might occur because there are too many chunks compared to the number of images that actually need to be processed.
298 lines
12 KiB
Python
298 lines
12 KiB
Python
__version__ = "4.0"
|
|
|
|
import json
|
|
import os
|
|
from collections import Counter
|
|
|
|
from meshroom.core import desc
|
|
|
|
|
|
def findMetadata(d, keys, defaultValue):
|
|
v = None
|
|
for key in keys:
|
|
v = d.get(key, None)
|
|
k = key.lower()
|
|
if v is not None:
|
|
return v
|
|
for dk, dv in d.items():
|
|
dkm = dk.lower().replace(" ", "")
|
|
if dkm == key.lower():
|
|
return dv
|
|
dkm = dkm.split(":")[-1]
|
|
dkm = dkm.split("/")[-1]
|
|
if dkm == k:
|
|
return dv
|
|
return defaultValue
|
|
|
|
|
|
class DividedInputNodeSize(desc.DynamicNodeSize):
|
|
'''
|
|
The LDR2HDR will reduce the amount of views in the SfMData.
|
|
This class converts the number of LDR input views into the number of HDR output views.
|
|
'''
|
|
def __init__(self, param, divParam):
|
|
super(DividedInputNodeSize, self).__init__(param)
|
|
self._divParam = divParam
|
|
|
|
def computeSize(self, node):
|
|
s = super(DividedInputNodeSize, self).computeSize(node)
|
|
divParam = node.attribute(self._divParam)
|
|
if divParam.value == 0:
|
|
return s
|
|
# s is the total number of inputs and may include outliers, that will not be used
|
|
# during computations and should thus be excluded from the size computation
|
|
return (s - node.outliersNb) / divParam.value
|
|
|
|
|
|
class LdrToHdrSampling(desc.AVCommandLineNode):
|
|
commandLine = 'aliceVision_LdrToHdrSampling {allParams}'
|
|
size = DividedInputNodeSize('input', 'nbBrackets')
|
|
parallelization = desc.Parallelization(blockSize=2)
|
|
commandLineRange = '--rangeStart {rangeStart} --rangeSize {rangeBlockSize}'
|
|
|
|
category = 'Panorama HDR'
|
|
documentation = '''
|
|
Sample pixels from Low range images for HDR creation.
|
|
'''
|
|
|
|
outliersNb = 0 # Number of detected outliers among the input images
|
|
|
|
inputs = [
|
|
desc.File(
|
|
name="input",
|
|
label="SfMData",
|
|
description="Input SfMData file.",
|
|
value="",
|
|
uid=[0],
|
|
),
|
|
desc.IntParam(
|
|
name="userNbBrackets",
|
|
label="Number Of Brackets",
|
|
description="Number of exposure brackets per HDR image (0 for automatic detection).",
|
|
value=0,
|
|
range=(0, 15, 1),
|
|
uid=[],
|
|
group="user", # not used directly on the command line
|
|
),
|
|
desc.IntParam(
|
|
name="nbBrackets",
|
|
label="Automatic Nb Brackets",
|
|
description="Number of exposure brackets used per HDR image.\n"
|
|
"It is detected automatically from input Viewpoints metadata if 'userNbBrackets'\n"
|
|
"is 0, else it is equal to 'userNbBrackets'.",
|
|
value=0,
|
|
range=(0, 10, 1),
|
|
uid=[0],
|
|
group="bracketsParams"
|
|
),
|
|
desc.BoolParam(
|
|
name="byPass",
|
|
label="Bypass",
|
|
description="Bypass HDR creation and use the medium bracket as the source for the next steps.",
|
|
value=False,
|
|
uid=[0],
|
|
enabled= lambda node: node.nbBrackets.value != 1,
|
|
),
|
|
desc.ChoiceParam(
|
|
name="calibrationMethod",
|
|
label="Calibration Method",
|
|
description="Method used for camera calibration:\n"
|
|
" - Linear: Disable the calibration and assumes a linear Camera Response Function. If images are encoded in a known colorspace (like sRGB for JPEG), the images will be automatically converted to linear.\n"
|
|
" - Debevec: This is the standard method for HDR calibration.\n"
|
|
" - Grossberg: Based on learned database of cameras, it allows to reduce the CRF to few parameters while keeping all the precision.\n"
|
|
" - Laguerre: Simple but robust method estimating the minimal number of parameters.",
|
|
values=["linear", "debevec", "grossberg", "laguerre"],
|
|
value="debevec",
|
|
exclusive=True,
|
|
uid=[0],
|
|
enabled= lambda node: node.byPass.enabled and not node.byPass.value,
|
|
),
|
|
desc.IntParam(
|
|
name="channelQuantizationPower",
|
|
label="Channel Quantization Power",
|
|
description="Quantization level like 8 bits or 10 bits.",
|
|
value=10,
|
|
range=(8, 14, 1),
|
|
uid=[0],
|
|
advanced=True,
|
|
enabled= lambda node: node.byPass.enabled and not node.byPass.value,
|
|
),
|
|
desc.ChoiceParam(
|
|
name="workingColorSpace",
|
|
label="Working Color Space",
|
|
description="Allows you to choose the color space in which the data are processed.",
|
|
value="sRGB",
|
|
values=["sRGB", "Linear", "ACES2065-1", "ACEScg", "no_conversion"],
|
|
exclusive=True,
|
|
uid=[0],
|
|
enabled= lambda node: node.byPass.enabled and not node.byPass.value,
|
|
),
|
|
desc.IntParam(
|
|
name="blockSize",
|
|
label="Block Size",
|
|
description="Size of the image tile to extract a sample.",
|
|
value=256,
|
|
range=(8, 1024, 1),
|
|
uid=[0],
|
|
advanced=True,
|
|
enabled= lambda node: node.byPass.enabled and not node.byPass.value,
|
|
),
|
|
desc.IntParam(
|
|
name="radius",
|
|
label="Patch Radius",
|
|
description="Radius of the patch used to analyze the sample statistics.",
|
|
value=5,
|
|
range=(0, 10, 1),
|
|
uid=[0],
|
|
advanced=True,
|
|
enabled= lambda node: node.byPass.enabled and not node.byPass.value,
|
|
),
|
|
desc.IntParam(
|
|
name="maxCountSample",
|
|
label="Max Number Of Samples",
|
|
description="Maximum number of samples per image group.",
|
|
value=200,
|
|
range=(10, 1000, 10),
|
|
uid=[0],
|
|
advanced=True,
|
|
enabled= lambda node: node.byPass.enabled and not node.byPass.value,
|
|
),
|
|
desc.BoolParam(
|
|
name="debug",
|
|
label="Export Debug Files",
|
|
description="Export debug files to analyze the sampling strategy.",
|
|
value=False,
|
|
uid=[],
|
|
enabled= lambda node: node.byPass.enabled and not node.byPass.value,
|
|
),
|
|
desc.ChoiceParam(
|
|
name="verboseLevel",
|
|
label="Verbose Level",
|
|
description="Verbosity level (fatal, error, warning, info, debug, trace).",
|
|
value="info",
|
|
values=["fatal", "error", "warning", "info", "debug", "trace"],
|
|
exclusive=True,
|
|
uid=[],
|
|
)
|
|
]
|
|
|
|
outputs = [
|
|
desc.File(
|
|
name="output",
|
|
label="Folder",
|
|
description="Output path for the samples.",
|
|
value=desc.Node.internalFolder,
|
|
uid=[],
|
|
),
|
|
]
|
|
|
|
def processChunk(self, chunk):
|
|
if chunk.node.nbBrackets.value == 1:
|
|
return
|
|
# Trick to avoid sending --nbBrackets to the command line when the bracket detection is automatic.
|
|
# Otherwise, the AliceVision executable has no way of determining whether the bracket detection was automatic
|
|
# or if it was hard-set by the user.
|
|
self.commandLine = "aliceVision_LdrToHdrSampling {allParams}"
|
|
if chunk.node.userNbBrackets.value == chunk.node.nbBrackets.value:
|
|
self.commandLine += "{bracketsParams}"
|
|
super(LdrToHdrSampling, self).processChunk(chunk)
|
|
|
|
@classmethod
|
|
def update(cls, node):
|
|
if not isinstance(node.nodeDesc, cls):
|
|
raise ValueError("Node {} is not an instance of type {}".format(node, cls))
|
|
# TODO: use Node version for this test
|
|
if "userNbBrackets" not in node.getAttributes().keys():
|
|
# Old version of the node
|
|
return
|
|
node.outliersNb = 0 # Reset the number of detected outliers
|
|
if node.userNbBrackets.value != 0:
|
|
node.nbBrackets.value = node.userNbBrackets.value
|
|
return
|
|
cameraInitOutput = node.input.getLinkParam(recursive=True)
|
|
if not cameraInitOutput:
|
|
node.nbBrackets.value = 0
|
|
return
|
|
if not cameraInitOutput.node.hasAttribute("viewpoints"):
|
|
if cameraInitOutput.node.hasAttribute("input"):
|
|
cameraInitOutput = cameraInitOutput.node.input.getLinkParam(recursive=True)
|
|
if cameraInitOutput and cameraInitOutput.node and cameraInitOutput.node.hasAttribute("viewpoints"):
|
|
viewpoints = cameraInitOutput.node.viewpoints.value
|
|
else:
|
|
# No connected CameraInit
|
|
node.nbBrackets.value = 0
|
|
return
|
|
|
|
inputs = []
|
|
for viewpoint in viewpoints:
|
|
jsonMetadata = viewpoint.metadata.value
|
|
if not jsonMetadata:
|
|
# no metadata, we cannot find the number of brackets
|
|
node.nbBrackets.value = 0
|
|
return
|
|
d = json.loads(jsonMetadata)
|
|
fnumber = findMetadata(d, ["FNumber", "Exif:ApertureValue", "ApertureValue", "Aperture"], "")
|
|
shutterSpeed = findMetadata(d, ["Exif:ShutterSpeedValue", "ShutterSpeedValue", "ShutterSpeed"], "")
|
|
iso = findMetadata(d, ["Exif:ISOSpeedRatings", "ISOSpeedRatings", "ISO"], "")
|
|
if not fnumber and not shutterSpeed:
|
|
# If one image without shutter or fnumber, we cannot found the number of brackets.
|
|
# We assume that there is no multi-bracketing, so nothing to do.
|
|
node.nbBrackets.value = 1
|
|
return
|
|
inputs.append((viewpoint.path.value, (float(fnumber), float(shutterSpeed), float(iso))))
|
|
inputs.sort()
|
|
|
|
exposureGroups = []
|
|
exposures = []
|
|
prevFnumber = 0.0
|
|
prevShutterSpeed = 0.0
|
|
prevIso = 0.0
|
|
prevPath = None # Stores the dirname of the previous parsed image
|
|
newGroup = False # True if a new exposure group needs to be created (useful when there are several datasets)
|
|
for path, exp in inputs:
|
|
# If the dirname of the previous image and the dirname of the current image do not match, this means that the
|
|
# dataset that is being parsed has changed. A new group needs to be created but will fail to be detected in the
|
|
# next "if" statement if the new dataset's exposure levels are different. Setting "newGroup" to True prevents this
|
|
# from happening.
|
|
if prevPath is not None and prevPath != os.path.dirname(path):
|
|
newGroup = True
|
|
|
|
# A new group is created if the current image's exposure level is larger than the previous image's, if there
|
|
# were any changes in the ISO or aperture value, or if a new dataset has been detected with the path.
|
|
# Since the input images are ordered, the shutter speed should always be decreasing, so a shutter speed larger
|
|
# than the previous one indicates the start of a new exposure group.
|
|
fnumber, shutterSpeed, iso = exp
|
|
if exposures:
|
|
prevFnumber, prevShutterSpeed, prevIso = exposures[-1]
|
|
if exposures and len(exposures) > 1 and (fnumber != prevFnumber or shutterSpeed > prevShutterSpeed or iso != prevIso) or newGroup:
|
|
exposureGroups.append(exposures)
|
|
exposures = [exp]
|
|
else:
|
|
exposures.append(exp)
|
|
|
|
prevPath = os.path.dirname(path)
|
|
newGroup = False
|
|
|
|
exposureGroups.append(exposures)
|
|
|
|
exposures = None
|
|
bracketSizes = Counter()
|
|
if len(exposureGroups) == 1:
|
|
if len(set(exposureGroups[0])) == 1:
|
|
# Single exposure and multiple views
|
|
node.nbBrackets.value = 1
|
|
else:
|
|
# Single view and multiple exposures
|
|
node.nbBrackets.value = len(exposureGroups[0])
|
|
else:
|
|
for expGroup in exposureGroups:
|
|
bracketSizes[len(expGroup)] += 1
|
|
|
|
if len(bracketSizes) == 0:
|
|
node.nbBrackets.value = 0
|
|
else:
|
|
bestTuple = bracketSizes.most_common(1)[0]
|
|
bestBracketSize = bestTuple[0]
|
|
bestCount = bestTuple[1]
|
|
node.outliersNb = len(inputs) - (bestBracketSize * bestCount) # Compute number of outliers
|
|
node.nbBrackets.value = bestBracketSize
|