forked from bartvdbraak/blender
949 lines
32 KiB
Python
949 lines
32 KiB
Python
#!BPY
|
|
|
|
"""
|
|
Name: 'Wavefront (.obj)...'
|
|
Blender: 237
|
|
Group: 'Import'
|
|
Tooltip: 'Load a Wavefront OBJ File, Shift: batch import all dir.'
|
|
"""
|
|
|
|
__author__ = "Campbell Barton"
|
|
__url__ = ["blender", "elysiun"]
|
|
__version__ = "1.0"
|
|
|
|
__bpydoc__ = """\
|
|
This script imports OBJ files to Blender.
|
|
|
|
Usage:
|
|
|
|
Run this script from "File->Import" menu and then load the desired OBJ file.
|
|
"""
|
|
|
|
# $Id$
|
|
#
|
|
# --------------------------------------------------------------------------
|
|
# OBJ Import v1.0 by Campbell Barton (AKA Ideasman)
|
|
# --------------------------------------------------------------------------
|
|
# ***** BEGIN GPL LICENSE BLOCK *****
|
|
#
|
|
# This program is free software; you can redistribute it and/or
|
|
# modify it under the terms of the GNU General Public License
|
|
# as published by the Free Software Foundation; either version 2
|
|
# of the License, or (at your option) any later version.
|
|
#
|
|
# This program is distributed in the hope that it will be useful,
|
|
# but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
# GNU General Public License for more details.
|
|
#
|
|
# You should have received a copy of the GNU General Public License
|
|
# along with this program; if not, write to the Free Software Foundation,
|
|
# Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
|
|
#
|
|
# ***** END GPL LICENCE BLOCK *****
|
|
# --------------------------------------------------------------------------
|
|
|
|
|
|
#==============================================#
|
|
# Return directory, where the file is #
|
|
#==============================================#
|
|
def stripFile(path):
|
|
lastSlash = max(path.rfind('\\'), path.rfind('/'))
|
|
if lastSlash != -1:
|
|
path = path[:lastSlash]
|
|
return '%s%s' % (path, sys.sep)
|
|
|
|
#==============================================#
|
|
# Strips the slashes from the back of a string #
|
|
#==============================================#
|
|
def stripPath(path):
|
|
return path.split('/')[-1].split('\\')[-1]
|
|
|
|
#====================================================#
|
|
# Strips the prefix off the name before writing #
|
|
#====================================================#
|
|
def stripExt(name): # name is a string
|
|
index = name.rfind('.')
|
|
if index != -1:
|
|
return name[ : index ]
|
|
else:
|
|
return name
|
|
|
|
|
|
from Blender import *
|
|
|
|
|
|
|
|
# Adds a slash to the end of a path if its not there.
|
|
def addSlash(path):
|
|
if path.endswith('\\') or path.endswith('/'):
|
|
return path
|
|
return path + sys.sep
|
|
|
|
|
|
def getExt(name):
|
|
index = name.rfind('.')
|
|
if index != -1:
|
|
return name[index+1:]
|
|
return name
|
|
|
|
try:
|
|
import os
|
|
except:
|
|
# So we know if os exists.
|
|
print 'Module "os" not found, install python to enable comprehensive image finding and batch loading.'
|
|
os = None
|
|
|
|
#===========================================================================#
|
|
# Comprehansive image loader, will search and find the image #
|
|
# Will return a blender image or none if the image is missing #
|
|
#===========================================================================#
|
|
def comprehansiveImageLoad(imagePath, filePath):
|
|
|
|
# When we have the file load it with this. try/except niceness.
|
|
def imageLoad(path):
|
|
try:
|
|
img = Image.Load(path)
|
|
print '\t\tImage loaded "%s"' % path
|
|
return img
|
|
except:
|
|
print '\t\tImage failed loading "%s", mabe its not a format blender can read.' % (path)
|
|
return None
|
|
|
|
# Image formats blender can read
|
|
IMAGE_EXT = ['jpg', 'jpeg', 'png', 'tga', 'bmp', 'rgb', 'sgi', 'bw', 'iff', 'lbm', # Blender Internal
|
|
'gif', 'psd', 'tif', 'tiff', 'pct', 'pict', 'pntg', 'qtif'] # Quacktime, worth a try.
|
|
|
|
|
|
|
|
|
|
print '\tAttempting to load "%s"' % imagePath
|
|
if sys.exists(imagePath):
|
|
print '\t\tFile found where expected.'
|
|
return imageLoad(imagePath)
|
|
|
|
imageFileName = stripPath(imagePath) # image path only
|
|
imageFileName_lower = imageFileName.lower() # image path only
|
|
imageFileName_noext = stripExt(imageFileName) # With no extension.
|
|
imageFileName_noext_lower = stripExt(imageFileName_lower) # With no extension.
|
|
imageFilePath = stripFile(imagePath)
|
|
|
|
# Remove relative path from image path
|
|
if imageFilePath.startswith('./') or imageFilePath.startswith('.\\'):
|
|
imageFilePath = imageFilePath[2:]
|
|
|
|
|
|
# Attempt to load from obj path.
|
|
tmpPath = stripFile(filePath) + stripFile(imageFilePath)
|
|
if sys.exists(tmpPath):
|
|
print '\t\tFile found in obj dir.'
|
|
return imageLoad(imagePath)
|
|
|
|
# OS NEEDED IF WE GO ANY FURTHER.
|
|
if not os:
|
|
return
|
|
|
|
|
|
# We have os.
|
|
# GATHER PATHS.
|
|
paths = {} # Store possible paths we may use, dict for no doubles.
|
|
tmpPath = addSlash(sys.expandpath('//')) # Blenders path
|
|
if sys.exists(tmpPath):
|
|
print '\t\tSearching in %s' % tmpPath
|
|
paths[tmpPath] = [os.listdir(tmpPath)] # Orig name for loading
|
|
paths[tmpPath].append([f.lower() for f in paths[tmpPath][0]]) # Lower case list.
|
|
paths[tmpPath].append([stripExt(f) for f in paths[tmpPath][1]]) # Lower case no ext
|
|
|
|
tmpPath = imageFilePath
|
|
if sys.exists(tmpPath):
|
|
print '\t\tSearching in %s' % tmpPath
|
|
paths[tmpPath] = [os.listdir(tmpPath)] # Orig name for loading
|
|
paths[tmpPath].append([f.lower() for f in paths[tmpPath][0]]) # Lower case list.
|
|
paths[tmpPath].append([stripExt(f) for f in paths[tmpPath][1]]) # Lower case no ext
|
|
|
|
tmpPath = stripFile(filePath)
|
|
if sys.exists(tmpPath):
|
|
print '\t\tSearching in %s' % tmpPath
|
|
paths[tmpPath] = [os.listdir(tmpPath)] # Orig name for loading
|
|
paths[tmpPath].append([f.lower() for f in paths[tmpPath][0]]) # Lower case list.
|
|
paths[tmpPath].append([stripExt(f) for f in paths[tmpPath][1]]) # Lower case no ext
|
|
|
|
tmpPath = addSlash(Get('texturesdir'))
|
|
if tmpPath and sys.exists(tmpPath):
|
|
print '\t\tSearching in %s' % tmpPath
|
|
paths[tmpPath] = [os.listdir(tmpPath)] # Orig name for loading
|
|
paths[tmpPath].append([f.lower() for f in paths[tmpPath][0]]) # Lower case list.
|
|
paths[tmpPath].append([stripExt(f) for f in paths[tmpPath][1]]) # Lower case no ext
|
|
|
|
# Add path if relative image patrh was given.
|
|
for k in paths.iterkeys():
|
|
tmpPath = k + imageFilePath
|
|
if sys.exists(tmpPath):
|
|
paths[tmpPath] = [os.listdir(tmpPath)] # Orig name for loading
|
|
paths[tmpPath].append([f.lower() for f in paths[tmpPath][0]]) # Lower case list.
|
|
paths[tmpPath].append([stripExt(f) for f in paths[tmpPath][1]]) # Lower case no ext
|
|
# DONE
|
|
|
|
|
|
#
|
|
for path, files in paths.iteritems():
|
|
|
|
if sys.exists(path + imageFileName):
|
|
return imageLoad(path + imageFileName)
|
|
|
|
# If the files not there then well do a case insensitive seek.
|
|
filesOrigCase = files[0]
|
|
filesLower = files[1]
|
|
filesLowerNoExt = files[2]
|
|
|
|
# We are going to try in index the file directly, if its not there just keep on
|
|
index = None
|
|
try:
|
|
# Is it just a case mismatch?
|
|
index = filesLower.index(imageFileName_lower)
|
|
except:
|
|
try:
|
|
# Have the extensions changed?
|
|
index = filesLowerNoExt.index(imageFileName_noext_lower)
|
|
|
|
ext = getExt( filesLower[index] ) # Get the extension of the file that matches all but ext.
|
|
|
|
# Check that the ext is useable eg- not a 3ds file :)
|
|
if ext.lower() not in IMAGE_EXT:
|
|
index = None
|
|
|
|
except:
|
|
index = None
|
|
|
|
if index != None:
|
|
tmpPath = path + filesOrigCase[index]
|
|
img = imageLoad( tmpPath )
|
|
if img != None:
|
|
print '\t\tImage Found "%s"' % tmpPath
|
|
return img
|
|
|
|
|
|
# IMAGE NOT FOUND IN ANY OF THE DIRS!, DO A RECURSIVE SEARCH.
|
|
print '\t\tImage Not Found in any of the dirs, doing a recusrive search'
|
|
for path in paths.iterkeys():
|
|
# Were not going to use files
|
|
|
|
|
|
#------------------
|
|
# finds the file starting at the root.
|
|
# def findImage(findRoot, imagePath):
|
|
#W---------------
|
|
|
|
# ROOT, DIRS, FILES
|
|
pathWalk = os.walk(path)
|
|
pathList = [True]
|
|
|
|
matchList = [] # Store a list of (match, size), choose the biggest.
|
|
while True:
|
|
try:
|
|
pathList = pathWalk.next()
|
|
except:
|
|
break
|
|
|
|
for file in pathList[2]:
|
|
file_lower = file.lower()
|
|
# FOUND A MATCH
|
|
if (file_lower == imageFileName_lower) or\
|
|
(stripExt(file_lower) == imageFileName_noext_lower and getExt(file_lower) in IMAGE_EXT):
|
|
name = pathList[0] + sys.sep + file
|
|
size = os.path.getsize(name)
|
|
print '\t\t\tfound:', name
|
|
matchList.append( (name, size) )
|
|
|
|
if matchList:
|
|
# Sort by file size
|
|
matchList.sort(lambda A, B: cmp(B[1], A[1]) )
|
|
|
|
print '\t\tFound "%s"' % matchList[0][0]
|
|
|
|
# Loop through all we have found
|
|
img = None
|
|
for match in matchList:
|
|
img = imageLoad(match[0]) # 0 - first, 0 - pathname
|
|
if img != None:
|
|
break
|
|
return img
|
|
|
|
# No go.
|
|
print '\t\tImage Not Found "%s"' % imagePath
|
|
return None
|
|
|
|
|
|
|
|
|
|
|
|
#==================================================================================#
|
|
# This function sets textures defined in .mtl file #
|
|
#==================================================================================#
|
|
# ___ Replaced by comprehensive imahge get
|
|
|
|
#==================================================================================#
|
|
# This function sets textures defined in .mtl file #
|
|
#==================================================================================#
|
|
def loadMaterialImage(mat, img_fileName, type, meshDict, dir):
|
|
TEX_ON_FLAG = NMesh.FaceModes['TEX']
|
|
|
|
texture = Texture.New(type)
|
|
texture.setType('Image')
|
|
|
|
# Absolute path - c:\.. etc would work here
|
|
image = comprehansiveImageLoad(img_fileName, dir)
|
|
|
|
if image:
|
|
texture.image = image
|
|
|
|
# adds textures to faces (Textured/Alt-Z mode)
|
|
# Only apply the diffuse texture to the face if the image has not been set with the inline usemat func.
|
|
if image and type == 'Kd':
|
|
for meshPair in meshDict.itervalues():
|
|
for f in meshPair[0].faces:
|
|
#print meshPair[0].materials[f.mat].name, mat.name
|
|
if meshPair[0].materials[f.mat].name == mat.name:
|
|
# the inline usemat command overides the material Image
|
|
if not f.image:
|
|
f.mode |= TEX_ON_FLAG
|
|
f.image = image
|
|
|
|
# adds textures for materials (rendering)
|
|
elif type == 'Ka':
|
|
mat.setTexture(0, texture, Texture.TexCo.UV, Texture.MapTo.CMIR) # TODO- Add AMB to BPY API
|
|
elif type == 'Kd':
|
|
mat.setTexture(1, texture, Texture.TexCo.UV, Texture.MapTo.COL)
|
|
elif type == 'Ks':
|
|
mat.setTexture(2, texture, Texture.TexCo.UV, Texture.MapTo.SPEC)
|
|
|
|
elif type == 'Bump': # New Additions
|
|
mat.setTexture(3, texture, Texture.TexCo.UV, Texture.MapTo.NOR)
|
|
elif type == 'D':
|
|
mat.setTexture(4, texture, Texture.TexCo.UV, Texture.MapTo.ALPHA)
|
|
elif type == 'refl':
|
|
mat.setTexture(5, texture, Texture.TexCo.UV, Texture.MapTo.REF)
|
|
|
|
|
|
#==================================================================================#
|
|
# This function loads materials from .mtl file (have to be defined in obj file) #
|
|
#==================================================================================#
|
|
def load_mtl(dir, mtl_file, meshDict, materialDict):
|
|
|
|
#===============================================================================#
|
|
# This gets a mat or creates one of the requested name if none exist. #
|
|
#===============================================================================#
|
|
def getMat(matName, materialDict):
|
|
# Make a new mat
|
|
try:
|
|
return materialDict[matName]
|
|
#except NameError or KeyError:
|
|
except: # Better do any exception
|
|
# Do we realy need to keep the dict up to date?, not realy but keeps consuistant.
|
|
materialDict[matName] = Material.New(matName)
|
|
return materialDict[matName]
|
|
|
|
|
|
mtl_file = stripPath(mtl_file)
|
|
mtl_fileName = dir + mtl_file
|
|
|
|
try:
|
|
fileLines= open(mtl_fileName, 'r').readlines()
|
|
except IOError:
|
|
print '\tunable to open referenced material file: "%s"' % mtl_fileName
|
|
return
|
|
|
|
try:
|
|
lIdx=0
|
|
while lIdx < len(fileLines):
|
|
l = fileLines[lIdx].split()
|
|
|
|
# Detect a line that will be ignored
|
|
if len(l) == 0 or l[0].startswith('#'):
|
|
pass
|
|
elif l[0] == 'newmtl':
|
|
currentMat = getMat('_'.join(l[1:]), materialDict) # Material should alredy exist.
|
|
elif l[0] == 'Ka':
|
|
currentMat.setMirCol((float(l[1]), float(l[2]), float(l[3])))
|
|
elif l[0] == 'Kd':
|
|
currentMat.setRGBCol((float(l[1]), float(l[2]), float(l[3])))
|
|
elif l[0] == 'Ks':
|
|
currentMat.setSpecCol((float(l[1]), float(l[2]), float(l[3])))
|
|
elif l[0] == 'Ns':
|
|
currentMat.setHardness( int((float(l[1])*0.51)) )
|
|
elif l[0] == 'Ni': # Refraction index
|
|
currentMat.setIOR( max(1, min(float(l[1]), 3))) # Between 1 and 3
|
|
elif l[0] == 'd':
|
|
currentMat.setAlpha(float(l[1]))
|
|
elif l[0] == 'Tr':
|
|
currentMat.setAlpha(float(l[1]))
|
|
elif l[0] == 'map_Ka':
|
|
img_fileName = ' '.join(l[1:])
|
|
loadMaterialImage(currentMat, img_fileName, 'Ka', meshDict, dir)
|
|
elif l[0] == 'map_Ks':
|
|
img_fileName = ' '.join(l[1:])
|
|
loadMaterialImage(currentMat, img_fileName, 'Ks', meshDict, dir)
|
|
elif l[0] == 'map_Kd':
|
|
img_fileName = ' '.join(l[1:])
|
|
loadMaterialImage(currentMat, img_fileName, 'Kd', meshDict, dir)
|
|
|
|
# new additions
|
|
elif l[0] == 'map_Bump': # Bumpmap
|
|
img_fileName = ' '.join(l[1:])
|
|
loadMaterialImage(currentMat, img_fileName, 'Bump', meshDict, dir)
|
|
elif l[0] == 'map_D': # Alpha map - Dissolve
|
|
img_fileName = ' '.join(l[1:])
|
|
loadMaterialImage(currentMat, img_fileName, 'D', meshDict, dir)
|
|
|
|
elif l[0] == 'refl': # Reflectionmap
|
|
img_fileName = ' '.join(l[1:])
|
|
loadMaterialImage(currentMat, img_fileName, 'refl', meshDict, dir)
|
|
|
|
lIdx+=1
|
|
except:
|
|
print '\tERROR: Unable to parse MTL file: "%s"' % mtl_file
|
|
return
|
|
print '\tUsing MTL: "%s"' % mtl_file
|
|
#===========================================================================#
|
|
# Returns unique name of object/mesh (preserve overwriting existing meshes) #
|
|
#===========================================================================#
|
|
def getUniqueName(name):
|
|
newName = name[:19] # 19 chars is the longest name.
|
|
uniqueInt = 0
|
|
while newName in getUniqueName.uniqueNames:
|
|
newName = '%s.%.3i' % (name[:15], uniqueInt)
|
|
uniqueInt +=1
|
|
getUniqueName.uniqueNames.append(newName)
|
|
return newName
|
|
getUniqueName.uniqueNames = []
|
|
|
|
#==================================================================================#
|
|
# This loads data from .obj file #
|
|
#==================================================================================#
|
|
def load_obj(file, IMPORT_MTL=1, IMPORT_EDGES=1, IMPORT_SMOOTH_ALL=0):
|
|
|
|
print '\nImporting OBJ file: "%s"' % file
|
|
|
|
time1 = sys.time()
|
|
|
|
getUniqueName.uniqueNames.extend( [ob.name for ob in Object.Get()] )
|
|
getUniqueName.uniqueNames.extend( NMesh.GetNames() )
|
|
|
|
# Deselect all objects in the scene.
|
|
# do this first so we dont have to bother, with objects we import
|
|
for ob in Scene.GetCurrent().getChildren():
|
|
ob.sel = 0
|
|
|
|
TEX_OFF_FLAG = ~NMesh.FaceModes['TEX']
|
|
|
|
# Get the file name with no path or .obj
|
|
fileName = stripExt( stripPath(file) )
|
|
|
|
mtl_fileName = [] # Support multiple mtl files if needed.
|
|
|
|
DIR = stripFile(file)
|
|
|
|
tempFile = open(file, 'r')
|
|
fileLines = tempFile.readlines()
|
|
tempFile.close()
|
|
del tempFile
|
|
uvMapList = [] # store tuple uv pairs here
|
|
|
|
# This dummy vert makes life a whole lot easier-
|
|
# pythons index system then aligns with objs, remove later
|
|
vertList = [] # Could havea vert but since this is a placeholder theres no Point
|
|
|
|
|
|
# Store all imported images in a dict, names are key
|
|
imageDict = {}
|
|
|
|
# This stores the index that the current mesh has for the current material.
|
|
# if the mesh does not have the material then set -1
|
|
contextMeshMatIdx = -1
|
|
|
|
# Keep this out of the dict for easy accsess.
|
|
nullMat = Material.New('(null)')
|
|
|
|
currentMat = nullMat # Use this mat.
|
|
currentImg = None # Null image is a string, otherwise this should be set to an image object.\
|
|
if IMPORT_SMOOTH_ALL:
|
|
currentSmooth = True
|
|
else:
|
|
currentSmooth = False
|
|
|
|
# Store a list of unnamed names
|
|
currentUnnamedGroupIdx = 1
|
|
currentUnnamedObjectIdx = 1
|
|
|
|
quadList = (0, 1, 2, 3)
|
|
|
|
faceQuadVList = [None, None, None, None]
|
|
faceTriVList = [None, None, None]
|
|
|
|
#==================================================================================#
|
|
# Load all verts first (texture verts too) #
|
|
#==================================================================================#
|
|
|
|
print '\tfile length: %d' % len(fileLines)
|
|
|
|
# Ignore normals and comments.
|
|
fileLines = [lsplit for l in fileLines if not l.startswith('vn') if not l.startswith('#') for lsplit in (l.split(),) if lsplit]
|
|
Vert = NMesh.Vert
|
|
vertList = [Vert(float(l[1]), float(l[2]), float(l[3]) ) for l in fileLines if l[0] == 'v']
|
|
uvMapList = [(float(l[1]), float(l[2])) for l in fileLines if l[0] == 'vt']
|
|
smoothingGroups = dict([('_'.join(l[1:]), None) for l in fileLines if l[0] == 's' ])
|
|
materialDict = dict([('_'.join(l[1:]), None) for l in fileLines if l[0] == 'usemtl']) # Store all imported materials as unique dict, names are key
|
|
print '\tvert:%i texverts:%i smoothgroups:%i materials:%s' % (len(vertList), len(uvMapList), len(smoothingGroups), len(materialDict))
|
|
|
|
# Replace filelines, Excluding v excludes "v ", "vn " and "vt "
|
|
|
|
# Remove any variables we may have created.
|
|
try: del _dummy
|
|
except: pass
|
|
try: del _x
|
|
except: pass
|
|
try: del _y
|
|
except: pass
|
|
try: del _z
|
|
except: pass
|
|
try: del lsplit
|
|
except: pass
|
|
del Vert
|
|
|
|
# With negative values this is used a lot. make faster access.
|
|
len_uvMapList = len(uvMapList)
|
|
len_vertList = len(vertList)
|
|
|
|
# Only want unique keys anyway
|
|
smoothingGroups['(null)'] = None # Make sure we have at least 1.
|
|
smoothingGroups = smoothingGroups.keys()
|
|
print '\tfound %d smoothing groups.' % (len(smoothingGroups) -1)
|
|
|
|
# Add materials to Blender for later is in teh OBJ
|
|
for k in materialDict.iterkeys():
|
|
materialDict[k] = Material.New(k)
|
|
|
|
|
|
# Make a list of all unused vert indicies that we can copy from
|
|
VERT_USED_LIST = [0]*len_vertList
|
|
|
|
# Here we store a boolean list of which verts are used or not
|
|
# no we know weather to add them to the current mesh
|
|
# This is an issue with global vertex indicies being translated to per mesh indicies
|
|
# like blenders, we start with a dummy just like the vert.
|
|
# -1 means unused, any other value refers to the local mesh index of the vert.
|
|
|
|
# currentObjectName has a char in front of it that determins weather its a group or object.
|
|
# We ignore it when naming the object.
|
|
currentObjectName = 'unnamed_obj_0' # If we cant get one, use this
|
|
|
|
#meshDict = {} # The 3 variables below are stored in a tuple within this dict for each mesh
|
|
currentMesh = NMesh.GetRaw() # The NMesh representation of the OBJ group/Object
|
|
#currentUsedVertList = {} # A Dict of smooth groups, each smooth group has a list of used verts and they are generated on demand so as to save memory.
|
|
currentMaterialMeshMapping = {} # Used to store material indicies so we dont have to search the mesh for materials every time.
|
|
|
|
# Every mesh has a null smooth group, this is used if there are no smooth groups in the OBJ file.
|
|
# and when for faces where no smooth group is used.
|
|
currentSmoothGroup = '(null)' # The Name of the current smooth group
|
|
|
|
# For direct accsess to the Current Meshes, Current Smooth Groups- Used verts.
|
|
# This is of course context based and changes on the fly.
|
|
# Set the initial '(null)' Smooth group, every mesh has one.
|
|
currentUsedVertListSmoothGroup = VERT_USED_LIST[:]
|
|
currentUsedVertList= {currentSmoothGroup: currentUsedVertListSmoothGroup }
|
|
|
|
|
|
# 0:NMesh, 1:SmoothGroups[UsedVerts[0,0,0,0]], 2:materialMapping['matname':matIndexForThisNMesh]
|
|
meshDict = {currentObjectName: (currentMesh, currentUsedVertList, currentMaterialMeshMapping) }
|
|
|
|
# Only show the bad uv error once
|
|
badObjUvs = 0
|
|
badObjFaceVerts = 0
|
|
badObjFaceTexCo = 0
|
|
|
|
|
|
#currentMesh.verts.append(vertList[0]) # So we can sync with OBJ indicies where 1 is the first item.
|
|
if len_uvMapList > 1:
|
|
currentMesh.hasFaceUV(1) # Turn UV's on if we have ANY texture coords in this obj file.
|
|
|
|
|
|
#==================================================================================#
|
|
# Load all faces into objects, main loop #
|
|
#==================================================================================#
|
|
#lIdx = 0
|
|
# Face and Object loading LOOP
|
|
#while lIdx < len(fileLines):
|
|
# l = fileLines[lIdx]
|
|
#for lIdx
|
|
for l in fileLines:
|
|
if len(l) == 0:
|
|
continue
|
|
# FACE
|
|
elif l[0] == 'f':
|
|
# Make a face with the correct material.
|
|
|
|
# Add material to mesh
|
|
if contextMeshMatIdx == -1:
|
|
tmpMatLs = currentMesh.materials
|
|
|
|
if len(tmpMatLs) == 16:
|
|
contextMeshMatIdx = 0 # Use first material
|
|
print 'material overflow, attempting to use > 16 materials. defaulting to first.'
|
|
else:
|
|
contextMeshMatIdx = len(tmpMatLs)
|
|
currentMaterialMeshMapping[currentMat.name] = contextMeshMatIdx
|
|
currentMesh.addMaterial(currentMat)
|
|
|
|
# Set up vIdxLs : Verts
|
|
# Set up vtIdxLs : UV
|
|
# Start with a dummy objects so python accepts OBJs 1 is the first index.
|
|
vIdxLs = []
|
|
vtIdxLs = []
|
|
|
|
|
|
fHasUV = len_uvMapList # Assume the face has a UV until it sho it dosent, if there are no UV coords then this will start as 0.
|
|
for v in l[1:]:
|
|
# OBJ files can have // or / to seperate vert/texVert/normal
|
|
# this is a bit of a pain but we must deal with it.
|
|
objVert = v.split('/')
|
|
|
|
# Vert Index - OBJ supports negative index assignment (like python)
|
|
index = int(objVert[0])-1
|
|
# Account for negative indicies.
|
|
if index < 0:
|
|
index = len_vertList+index+1
|
|
|
|
vIdxLs.append(index)
|
|
if fHasUV:
|
|
# UV
|
|
index = 0 # Dummy var
|
|
if len(objVert) == 1:
|
|
index = vIdxLs[-1]
|
|
elif objVert[1]: # != '' # Its possible that theres no texture vert just he vert and normal eg 1//2
|
|
index = int(objVert[1])-1
|
|
if index < 0:
|
|
index = len_uvMapList+index+1
|
|
|
|
if len_uvMapList > index:
|
|
vtIdxLs.append(index) # Seperate UV coords
|
|
else:
|
|
# BAD FILE, I have found this so I account for it.
|
|
# INVALID UV COORD
|
|
# Could ignore this- only happens with 1 in 1000 files.
|
|
badObjFaceTexCo +=1
|
|
vtIdxLs.append(1)
|
|
|
|
fHasUV = 0
|
|
|
|
# Dont add a UV to the face if its larger then the UV coord list
|
|
# The OBJ file would have to be corrupt or badly written for thi to happen
|
|
# but account for it anyway.
|
|
if len(vtIdxLs) > 0:
|
|
if vtIdxLs[-1] > len_uvMapList:
|
|
fHasUV = 0
|
|
|
|
badObjUvs +=1 # ERROR, Cont
|
|
# Quads only, we could import quads using the method below but it polite to import a quad as a quad.
|
|
#print lIdx, len(vIdxLs), len(currentUsedVertListSmoothGroup)
|
|
#print fileLines[lIdx]
|
|
if len(vIdxLs) == 2:
|
|
if IMPORT_EDGES:
|
|
# Edge
|
|
for i in (0,1):
|
|
if currentUsedVertListSmoothGroup[vIdxLs[i]] == 0:
|
|
faceQuadVList[i] = vertList[vIdxLs[i]]
|
|
currentMesh.verts.append(faceQuadVList[i])
|
|
currentUsedVertListSmoothGroup[vIdxLs[i]] = len(currentMesh.verts)-1
|
|
else:
|
|
faceQuadVList[i] = currentMesh.verts[currentUsedVertListSmoothGroup[vIdxLs[i]]]
|
|
|
|
currentMesh.addEdge(faceQuadVList[0], faceQuadVList[1])
|
|
elif len(vIdxLs) == 4:
|
|
|
|
# Have found some files where wach face references the same vert
|
|
# - This causes a bug and stopts the import so lets check here
|
|
if vIdxLs[0] == vIdxLs[1] or\
|
|
vIdxLs[0] == vIdxLs[2] or\
|
|
vIdxLs[0] == vIdxLs[3] or\
|
|
vIdxLs[1] == vIdxLs[2] or\
|
|
vIdxLs[1] == vIdxLs[3] or\
|
|
vIdxLs[2] == vIdxLs[3]:
|
|
badObjFaceVerts+=1
|
|
else:
|
|
for i in quadList: # quadList == [0,1,2,3]
|
|
if currentUsedVertListSmoothGroup[vIdxLs[i]] == 0:
|
|
faceQuadVList[i] = vertList[vIdxLs[i]]
|
|
currentMesh.verts.append(faceQuadVList[i])
|
|
currentUsedVertListSmoothGroup[vIdxLs[i]] = len(currentMesh.verts)-1
|
|
else:
|
|
faceQuadVList[i] = currentMesh.verts[currentUsedVertListSmoothGroup[vIdxLs[i]]]
|
|
|
|
f = NMesh.Face(faceQuadVList)
|
|
# UV MAPPING
|
|
if fHasUV:
|
|
f.uv = [uvMapList[ vtIdxLs[0] ],uvMapList[ vtIdxLs[1] ],uvMapList[ vtIdxLs[2] ],uvMapList[ vtIdxLs[3] ]]
|
|
if currentImg:
|
|
f.image = currentImg
|
|
else:
|
|
f.mode &= TEX_OFF_FLAG
|
|
|
|
f.mat = contextMeshMatIdx
|
|
f.smooth = currentSmooth
|
|
currentMesh.faces.append(f) # move the face onto the mesh
|
|
|
|
elif len(vIdxLs) >= 3: # This handles tri's and fans
|
|
for i in range(len(vIdxLs)-2):
|
|
if vIdxLs[0] == vIdxLs[i+1] or\
|
|
vIdxLs[0] == vIdxLs[i+2] or\
|
|
vIdxLs[i+1] == vIdxLs[i+2]:
|
|
badObjFaceVerts+=1
|
|
else:
|
|
for k, j in [(0,0), (1,i+1), (2,i+2)]:
|
|
if currentUsedVertListSmoothGroup[vIdxLs[j]] == 0:
|
|
faceTriVList[k] = vertList[vIdxLs[j]]
|
|
currentMesh.verts.append(faceTriVList[k])
|
|
currentUsedVertListSmoothGroup[vIdxLs[j]] = len(currentMesh.verts)-1
|
|
else:
|
|
faceTriVList[k] = currentMesh.verts[currentUsedVertListSmoothGroup[vIdxLs[j]]]
|
|
|
|
f = NMesh.Face(faceTriVList)
|
|
|
|
# UV MAPPING
|
|
if fHasUV:
|
|
f.uv = [uvMapList[vtIdxLs[0]], uvMapList[vtIdxLs[i+1]], uvMapList[vtIdxLs[i+2]]]
|
|
if currentImg:
|
|
f.image = currentImg
|
|
else:
|
|
f.mode &= TEX_OFF_FLAG
|
|
|
|
f.mat = contextMeshMatIdx
|
|
f.smooth = currentSmooth
|
|
currentMesh.faces.append(f) # move the face onto the mesh
|
|
|
|
# FACE SMOOTHING
|
|
elif l[0] == 's':
|
|
# No value? then turn on.
|
|
if len(l) == 1:
|
|
currentSmooth = True
|
|
currentSmoothGroup = '(null)'
|
|
else:
|
|
if l[1] == 'off': # We all have a null group so dont need to try, will try anyway to avoid code duplication.
|
|
if not IMPORT_SMOOTH_ALL:
|
|
currentSmooth = False
|
|
currentSmoothGroup = '(null)'
|
|
else:
|
|
currentSmooth = True
|
|
currentSmoothGroup = '_'.join(l[1:])
|
|
try:
|
|
currentUsedVertListSmoothGroup = currentUsedVertList[currentSmoothGroup]
|
|
except KeyError:
|
|
currentUsedVertList[currentSmoothGroup] = currentUsedVertListSmoothGroup = VERT_USED_LIST[:]
|
|
|
|
|
|
# OBJECT / GROUP
|
|
elif l[0] == 'o' or l[0] == 'g':
|
|
|
|
# Forget about the current image
|
|
currentImg = None
|
|
|
|
# This makes sure that if an object and a group have the same name then
|
|
# they are not put into the same object.
|
|
|
|
# Only make a new group.object name if the verts in the existing object have been used, this is obscure
|
|
# but some files face groups seperating verts and faces which results in silly things. (no groups have names.)
|
|
if len(l) > 1:
|
|
currentObjectName = '_'.join(l[1:])
|
|
else: # No name given
|
|
# Make a new empty name
|
|
if l[0] == 'g': # Make a blank group name
|
|
currentObjectName = 'unnamed_grp_%.4d' % currentUnnamedGroupIdx
|
|
currentUnnamedGroupIdx +=1
|
|
else: # is an object.
|
|
currentObjectName = 'unnamed_ob_%.4d' % currentUnnamedObjectIdx
|
|
currentUnnamedObjectIdx +=1
|
|
|
|
|
|
# If we havnt written to this mesh before then do so.
|
|
# if we have then we'll just keep appending to it, this is required for soem files.
|
|
|
|
# If we are new, or we are not yet in the list of added meshes
|
|
# then make us new mesh.
|
|
if len(l) == 1 or (not meshDict.has_key(currentObjectName)):
|
|
currentMesh = NMesh.GetRaw()
|
|
|
|
currentUsedVertList = {}
|
|
|
|
# Sg is a string
|
|
########currentSmoothGroup = '(null)' # From examplesm changing the g/o shouldent change the smooth group.
|
|
currentUsedVertList[currentSmoothGroup] = currentUsedVertListSmoothGroup = VERT_USED_LIST[:]
|
|
|
|
currentMaterialMeshMapping = {}
|
|
meshDict[currentObjectName] = (currentMesh, currentUsedVertList, currentMaterialMeshMapping)
|
|
currentMesh.hasFaceUV(1)
|
|
contextMeshMatIdx = -1
|
|
|
|
else:
|
|
# Since we have this in Blender then we will check if the current Mesh has the material.
|
|
# set the contextMeshMatIdx to the meshs index but only if we have it.
|
|
currentMesh, currentUsedVertList, currentMaterialMeshMapping = meshDict[currentObjectName]
|
|
#getMeshMaterialIndex(currentMesh, currentMat)
|
|
|
|
try:
|
|
contextMeshMatIdx = currentMaterialMeshMapping[currentMat.name] #getMeshMaterialIndex(currentMesh, currentMat)
|
|
except KeyError:
|
|
contextMeshMatIdx -1
|
|
|
|
# For new meshes switch smoothing groups to null
|
|
########currentSmoothGroup = '(null)' # From examplesm changing the g/o shouldent change the smooth group.
|
|
try:
|
|
currentUsedVertListSmoothGroup = currentUsedVertList[currentSmoothGroup]
|
|
except:
|
|
currentUsedVertList[currentSmoothGroup] = currentUsedVertListSmoothGroup = VERT_USED_LIST[:]
|
|
|
|
|
|
|
|
# MATERIAL
|
|
elif l[0] == 'usemtl':
|
|
if len(l) == 1 or l[1] == '(null)':
|
|
currentMat = nullMat # We know we have a null mat.
|
|
else:
|
|
currentMat = materialDict['_'.join(l[1:])]
|
|
try:
|
|
contextMeshMatIdx = currentMaterialMeshMapping[currentMat.name]
|
|
except KeyError:
|
|
contextMeshMatIdx = -1 #getMeshMaterialIndex(currentMesh, currentMat)
|
|
|
|
# IMAGE
|
|
elif l[0] == 'usemat' or l[0] == 'usemap':
|
|
if len(l) == 1 or l[1] == '(null)' or l[1] == 'off':
|
|
currentImg = None
|
|
else:
|
|
# Load an image.
|
|
newImgName = stripPath(' '.join(l[1:])) # Use space since its a file name.
|
|
|
|
try:
|
|
# Assume its alredy set in the dict (may or maynot be loaded)
|
|
currentImg = imageDict[newImgName]
|
|
|
|
except KeyError: # Not in dict, add for first time.
|
|
# Image has not been added, Try and load the image
|
|
currentImg = comprehansiveImageLoad(newImgName, DIR) # Use join in case of spaces
|
|
imageDict[newImgName] = currentImg
|
|
# These may be None, thats okay.
|
|
|
|
|
|
|
|
# MATERIAL FILE
|
|
elif l[0] == 'mtllib':
|
|
mtl_fileName.append(' '.join(l[1:]) ) # SHOULD SUPPORT MULTIPLE MTL?
|
|
#lIdx+=1
|
|
|
|
# Applies material properties to materials alredy on the mesh as well as Textures.
|
|
if IMPORT_MTL:
|
|
for mtl in mtl_fileName:
|
|
load_mtl(DIR, mtl, meshDict, materialDict)
|
|
|
|
|
|
importedObjects = []
|
|
for mk, me in meshDict.iteritems():
|
|
nme = me[0]
|
|
|
|
# Ignore no vert meshes.
|
|
if not nme.verts: # == []
|
|
continue
|
|
|
|
name = getUniqueName(mk)
|
|
ob = NMesh.PutRaw(nme, name)
|
|
ob.name = name
|
|
|
|
importedObjects.append(ob)
|
|
|
|
# Select all imported objects.
|
|
for ob in importedObjects:
|
|
ob.sel = 1
|
|
if badObjUvs > 0:
|
|
print '\tERROR: found %d faces with badly formatted UV coords. everything else went okay.' % badObjUvs
|
|
|
|
if badObjFaceVerts > 0:
|
|
print '\tERROR: found %d faces reusing the same vertex. everything else went okay.' % badObjFaceVerts
|
|
|
|
if badObjFaceTexCo > 0:
|
|
print '\tERROR: found %d faces with invalit texture coords. everything else went okay.' % badObjFaceTexCo
|
|
|
|
|
|
print "obj import time: ", sys.time() - time1
|
|
|
|
def load_obj_ui(file):
|
|
|
|
IMPORT_MTL = Draw.Create(1)
|
|
IMPORT_DIR = Draw.Create(0)
|
|
IMPORT_NEW_SCENE = Draw.Create(0)
|
|
IMPORT_EDGES = Draw.Create(1)
|
|
IMPORT_SMOOTH_ALL = Draw.Create(0)
|
|
|
|
|
|
# Get USER Options
|
|
pup_block = [\
|
|
('Material (*.mtl)', IMPORT_MTL, 'Imports material settings and images from the obj\'s .mtl file'),\
|
|
('All *.obj\'s in dir', IMPORT_DIR, 'Import all obj files in this dir (avoid overlapping data with "Create scene")'),\
|
|
('Create scene', IMPORT_NEW_SCENE, 'Imports each obj into its own scene, named from the file'),\
|
|
'Geometry...',\
|
|
('Edges', IMPORT_EDGES, 'Import faces with 2 verts as in edge'),\
|
|
('Smooths all faces', IMPORT_SMOOTH_ALL, 'Smooth all faces even if they are not in a smoothing group'),\
|
|
]
|
|
|
|
if not os:
|
|
pup_block.pop(1) # Make sure this is the IMPORT_DIR option that requires OS
|
|
|
|
if not Draw.PupBlock('Import...', pup_block):
|
|
return
|
|
|
|
Window.WaitCursor(1)
|
|
Window.DrawProgressBar(0, '')
|
|
time = sys.time()
|
|
|
|
IMPORT_MTL = IMPORT_MTL.val
|
|
IMPORT_DIR = IMPORT_DIR.val
|
|
IMPORT_NEW_SCENE = IMPORT_NEW_SCENE.val
|
|
IMPORT_EDGES = IMPORT_EDGES.val
|
|
IMPORT_SMOOTH_ALL = IMPORT_SMOOTH_ALL.val
|
|
|
|
#orig_scene = Scene.GetCurrent()
|
|
|
|
obj_dir = stripFile(file)
|
|
if IMPORT_DIR:
|
|
obj_files = [(obj_dir,f) for f in os.listdir(obj_dir) if f.lower().endswith('obj')]
|
|
else:
|
|
obj_files = [(obj_dir,stripPath(file))]
|
|
|
|
obj_len = len(obj_files)
|
|
count = 0
|
|
for d, f in obj_files:
|
|
count+= 1
|
|
if not sys.exists(d+f):
|
|
print 'Error: "%s%s" does not exist' % (d,f)
|
|
else:
|
|
if IMPORT_NEW_SCENE:
|
|
scn = Scene.New('.'.join(f.split('.')[0:-1]))
|
|
scn.makeCurrent()
|
|
|
|
|
|
Window.DrawProgressBar((float(count)/obj_len) - 0.01, '%s: %i of %i' % (f, count, obj_len))
|
|
load_obj(d+f, IMPORT_MTL, IMPORT_EDGES, IMPORT_SMOOTH_ALL)
|
|
|
|
|
|
#orig_scene.makeCurrent() # We can leave them in there new scene.
|
|
Window.DrawProgressBar(1, '')
|
|
Window.WaitCursor(0)
|
|
|
|
if count > 1:
|
|
print 'Total obj import "%s" dir: %.2f' % (obj_dir, sys.time() - time)
|
|
|
|
|
|
def main():
|
|
Window.FileSelector(load_obj_ui, 'Import a Wavefront OBJ')
|
|
|
|
|
|
|
|
if __name__ == '__main__':
|
|
main() |