forked from bartvdbraak/blender
2dcf23c327
Added a new script - 'Dump All Images to Path' This script copies all the images used by 1 blend to a spesified path and references the new images from Blender Usefull for moving projects between computers or when you reference many images. naming collisions and multiple images using the same image path are delt with properly only creating new image names when needed. Blender images will reference the newly copied files - So be mindfull when you save your blend after running the script. Notes, images with the path "Untitled will be ignored" Image path collisions are managed by enumerating the path names so images will never be overwritten at the target path.
834 lines
25 KiB
Python
834 lines
25 KiB
Python
#!BPY
|
|
|
|
"""
|
|
Name: 'Wavefront (.obj)...'
|
|
Blender: 242
|
|
Group: 'Import'
|
|
Tooltip: 'Load a Wavefront OBJ File, Shift: batch import all dir.'
|
|
"""
|
|
|
|
__author__= "Campbell Barton", "Jiri Hnidek"
|
|
__url__= ["blender.org", "blenderartists.org"]
|
|
__version__= "2.0"
|
|
|
|
__bpydoc__= """\
|
|
This script imports a Wavefront OBJ files to Blender.
|
|
|
|
Usage:
|
|
Run this script from "File->Import" menu and then load the desired OBJ file.
|
|
Note, This loads mesh objects and materials only, nurbs and curves are not supported.
|
|
"""
|
|
|
|
from Blender import *
|
|
import BPyMesh
|
|
import BPyImage
|
|
import BPyMessages
|
|
|
|
try: import os
|
|
except: os= False
|
|
|
|
|
|
# Generic path functions
|
|
def stripFile(path):
|
|
'''Return directory, where the file is'''
|
|
lastSlash= max(path.rfind('\\'), path.rfind('/'))
|
|
if lastSlash != -1:
|
|
path= path[:lastSlash]
|
|
return '%s%s' % (path, sys.sep)
|
|
|
|
def stripPath(path):
|
|
'''Strips the slashes from the back of a string'''
|
|
return path.split('/')[-1].split('\\')[-1]
|
|
|
|
def stripExt(name): # name is a string
|
|
'''Strips the prefix off the name before writing'''
|
|
index= name.rfind('.')
|
|
if index != -1:
|
|
return name[ : index ]
|
|
else:
|
|
return name
|
|
# end path funcs
|
|
|
|
|
|
|
|
def line_value(line_split):
|
|
'''
|
|
Returns 1 string represneting the value for this line
|
|
None will be returned if theres only 1 word
|
|
'''
|
|
length= len(line_split)
|
|
if length == 1:
|
|
return None
|
|
|
|
elif length == 2:
|
|
return line_split[1]
|
|
|
|
elif length > 2:
|
|
return ' '.join( line_split[1:] )
|
|
|
|
def obj_image_load(imagepath, DIR, IMAGE_SEARCH):
|
|
'''
|
|
Mainly uses comprehensiveImageLoad
|
|
but tries to replace '_' with ' ' for Max's exporter replaces spaces with underscores.
|
|
'''
|
|
|
|
if '_' in imagepath:
|
|
image= BPyImage.comprehensiveImageLoad(imagepath, DIR, PLACE_HOLDER= False, RECURSIVE= IMAGE_SEARCH)
|
|
if image: return image
|
|
# Did the exporter rename the image?
|
|
image= BPyImage.comprehensiveImageLoad(imagepath.replace('_', ' '), DIR, PLACE_HOLDER= False, RECURSIVE= IMAGE_SEARCH)
|
|
if image: return image
|
|
|
|
# Return an image, placeholder if it dosnt exist
|
|
image= BPyImage.comprehensiveImageLoad(imagepath, DIR, PLACE_HOLDER= True, RECURSIVE= IMAGE_SEARCH)
|
|
return image
|
|
|
|
|
|
def create_materials(filepath, material_libs, unique_materials, unique_material_images, IMAGE_SEARCH):
|
|
'''
|
|
Create all the used materials in this obj,
|
|
assign colors and images to the materials from all referenced material libs
|
|
'''
|
|
DIR= stripFile(filepath)
|
|
|
|
#==================================================================================#
|
|
# This function sets textures defined in .mtl file #
|
|
#==================================================================================#
|
|
def load_material_image(blender_material, context_material_name, imagepath, type):
|
|
|
|
texture= Texture.New(type)
|
|
texture.setType('Image')
|
|
|
|
# Absolute path - c:\.. etc would work here
|
|
image= obj_image_load(imagepath, DIR, IMAGE_SEARCH)
|
|
|
|
texture.image = image
|
|
|
|
# Adds textures for materials (rendering)
|
|
if type == 'Kd':
|
|
if image.depth == 32:
|
|
# Image has alpha
|
|
blender_material.setTexture(0, texture, Texture.TexCo.UV, Texture.MapTo.COL | Texture.MapTo.ALPHA)
|
|
texture.setImageFlags('MipMap', 'InterPol', 'UseAlpha')
|
|
blender_material.mode |= Material.Modes.ZTRANSP
|
|
blender_material.alpha = 0.0
|
|
else:
|
|
blender_material.setTexture(0, texture, Texture.TexCo.UV, Texture.MapTo.COL)
|
|
|
|
# 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.
|
|
unique_material_images[context_material_name]= image # set the texface image
|
|
|
|
elif type == 'Ka':
|
|
blender_material.setTexture(1, texture, Texture.TexCo.UV, Texture.MapTo.CMIR) # TODO- Add AMB to BPY API
|
|
|
|
elif type == 'Ks':
|
|
blender_material.setTexture(2, texture, Texture.TexCo.UV, Texture.MapTo.SPEC)
|
|
|
|
elif type == 'Bump':
|
|
blender_material.setTexture(3, texture, Texture.TexCo.UV, Texture.MapTo.NOR)
|
|
elif type == 'D':
|
|
blender_material.setTexture(4, texture, Texture.TexCo.UV, Texture.MapTo.ALPHA)
|
|
blender_material.mode |= Material.Modes.ZTRANSP
|
|
blender_material.alpha = 0.0
|
|
# Todo, unset deffuse material alpha if it has an alpha channel
|
|
|
|
elif type == 'refl':
|
|
blender_material.setTexture(5, texture, Texture.TexCo.UV, Texture.MapTo.REF)
|
|
|
|
|
|
# Add an MTL with the same name as the obj if no MTLs are spesified.
|
|
temp_mtl= stripExt(stripPath(filepath))+ '.mtl'
|
|
|
|
if sys.exists(DIR + temp_mtl) and temp_mtl not in material_libs:
|
|
material_libs.append( temp_mtl )
|
|
del temp_mtl
|
|
|
|
#Create new materials
|
|
for name in unique_materials.iterkeys():
|
|
unique_materials[name]= Material.New(name)
|
|
|
|
unique_material_images[name]= None # assign None to all material images to start with, add to later.
|
|
|
|
unique_materials[None]= None
|
|
|
|
for libname in material_libs:
|
|
mtlpath= DIR + libname
|
|
if not sys.exists(mtlpath):
|
|
#print '\tError Missing MTL: "%s"' % mtlpath
|
|
pass
|
|
else:
|
|
#print '\t\tloading mtl: "%s"' % mtlpath
|
|
context_material= None
|
|
mtl= open(mtlpath)
|
|
for line in mtl.xreadlines():
|
|
if line.startswith('newmtl'):
|
|
context_material_name= line_value(line.split())
|
|
if unique_materials.has_key(context_material_name):
|
|
context_material = unique_materials[ context_material_name ]
|
|
else:
|
|
context_material = None
|
|
|
|
elif context_material:
|
|
# we need to make a material to assign properties to it.
|
|
line_split= line.split()
|
|
line_lower= line.lower()
|
|
if line_lower.startswith('ka'):
|
|
context_material.setMirCol((float(line_split[1]), float(line_split[2]), float(line_split[3])))
|
|
elif line_lower.startswith('kd'):
|
|
context_material.setRGBCol((float(line_split[1]), float(line_split[2]), float(line_split[3])))
|
|
elif line_lower.startswith('ks'):
|
|
context_material.setSpecCol((float(line_split[1]), float(line_split[2]), float(line_split[3])))
|
|
elif line_lower.startswith('ns'):
|
|
context_material.setHardness( int((float(line_split[1])*0.51)) )
|
|
elif line_lower.startswith('ni'): # Refraction index
|
|
context_material.setIOR( max(1, min(float(line_split[1]), 3))) # Between 1 and 3
|
|
elif line_lower.startswith('d') or line_lower.startswith('tr'):
|
|
context_material.setAlpha(float(line_split[1]))
|
|
elif line_lower.startswith('map_ka'):
|
|
img_filepath= line_value(line.split())
|
|
load_material_image(context_material, context_material_name, img_filepath, 'Ka')
|
|
elif line_lower.startswith('map_ks'):
|
|
img_filepath= line_value(line.split())
|
|
load_material_image(context_material, context_material_name, img_filepath, 'Ks')
|
|
elif line_lower.startswith('map_kd'):
|
|
img_filepath= line_value(line.split())
|
|
load_material_image(context_material, context_material_name, img_filepath, 'Kd')
|
|
elif line_lower.startswith('map_bump'):
|
|
img_filepath= line_value(line.split())
|
|
load_material_image(context_material, context_material_name, img_filepath, 'Bump')
|
|
elif line_lower.startswith('map_d') or line_lower.startswith('map_tr'): # Alpha map - Dissolve
|
|
img_filepath= line_value(line.split())
|
|
load_material_image(context_material, context_material_name, img_filepath, 'D')
|
|
|
|
elif line_lower.startswith('refl'): # Reflectionmap
|
|
img_filepath= line_value(line.split())
|
|
load_material_image(context_material, context_material_name, img_filepath, 'refl')
|
|
mtl.close()
|
|
|
|
|
|
def split_mesh(verts_loc, faces, unique_materials, SPLIT_OBJECTS, SPLIT_MATERIALS):
|
|
'''
|
|
Takes vert_loc and faces, and seperates into multiple sets of
|
|
(verts_loc, faces, unique_materials)
|
|
This is done so objects do not overload the 16 material limit.
|
|
'''
|
|
|
|
if not SPLIT_OBJECTS and not SPLIT_MATERIALS:
|
|
return [(verts_loc, faces, unique_materials)]
|
|
|
|
|
|
# Return a key that makes the faces unique.
|
|
if SPLIT_OBJECTS and not SPLIT_MATERIALS:
|
|
def face_key(face):
|
|
return face[4] # object
|
|
|
|
if not SPLIT_OBJECTS and SPLIT_MATERIALS:
|
|
def face_key(face):
|
|
return face[2] # material
|
|
|
|
else: # Both
|
|
def face_key(face):
|
|
return face[2], face[4] # material
|
|
|
|
|
|
face_split_dict= {}
|
|
|
|
oldkey= -1 # initialize to a value what will never match the key
|
|
|
|
for face in faces:
|
|
|
|
key= face_key(face)
|
|
|
|
if oldkey != key:
|
|
# Check the key has changed.
|
|
try:
|
|
faces_split, verts_split, unique_materials_split, vert_remap= face_split_dict[key]
|
|
except KeyError:
|
|
faces_split= []
|
|
verts_split= []
|
|
unique_materials_split= {}
|
|
vert_remap= [-1]*len(verts_loc)
|
|
|
|
face_split_dict[key]= (faces_split, verts_split, unique_materials_split, vert_remap)
|
|
|
|
oldkey= key
|
|
|
|
face_vert_loc_indicies= face[0]
|
|
|
|
# Remap verts to new vert list and add where needed
|
|
for enum, i in enumerate(face_vert_loc_indicies):
|
|
if vert_remap[i] == -1:
|
|
new_index= len(verts_split)
|
|
vert_remap[i]= new_index # set the new remapped index so we only add once and can reference next time.
|
|
face_vert_loc_indicies[enum] = new_index # remap to the local index
|
|
verts_split.append( verts_loc[i] ) # add the vert to the local verts
|
|
|
|
else:
|
|
face_vert_loc_indicies[enum] = vert_remap[i] # remap to the local index
|
|
|
|
matname= face[2]
|
|
if matname and not unique_materials_split.has_key(matname):
|
|
unique_materials_split[matname] = unique_materials[matname]
|
|
|
|
faces_split.append(face)
|
|
|
|
|
|
# remove one of the itemas and reorder
|
|
return [(verts_split, faces_split, unique_materials_split) for faces_split, verts_split, unique_materials_split, vert_remap in face_split_dict.itervalues()]
|
|
|
|
|
|
def create_mesh(new_objects, has_ngons, CREATE_FGONS, CREATE_EDGES, verts_loc, verts_tex, faces, unique_materials, unique_material_images, unique_smooth_groups):
|
|
'''
|
|
Takes all the data gathered and generates a mesh, adding the new object to new_objects
|
|
deals with fgons, sharp edges and assigning materials
|
|
'''
|
|
if not has_ngons:
|
|
CREATE_FGONS= False
|
|
|
|
if unique_smooth_groups:
|
|
sharp_edges= {}
|
|
smooth_group_users= dict([ (context_smooth_group, {}) for context_smooth_group in unique_smooth_groups.iterkeys() ])
|
|
context_smooth_group_old= -1
|
|
|
|
# Split fgons into tri's
|
|
fgon_edges= {} # Used for storing fgon keys
|
|
if CREATE_EDGES:
|
|
edges= []
|
|
|
|
context_object= None
|
|
|
|
# reverse loop through face indicies
|
|
for f_idx in xrange(len(faces)-1, -1, -1):
|
|
|
|
face_vert_loc_indicies,\
|
|
face_vert_tex_indicies,\
|
|
context_material,\
|
|
context_smooth_group,\
|
|
context_object= faces[f_idx]
|
|
|
|
len_face_vert_loc_indicies = len(face_vert_loc_indicies)
|
|
|
|
if len_face_vert_loc_indicies==1:
|
|
faces.pop(f_idx)# cant add single vert faces
|
|
|
|
elif not face_vert_tex_indicies: # images that are -1 are lines, a bit obscure but works.
|
|
if CREATE_EDGES:
|
|
try: # python 2.4+
|
|
edges.extend( (face_vert_loc_indicies[i], face_vert_loc_indicies[i+1]) for i in xrange(len_face_vert_loc_indicies-1) )
|
|
except: # python 2.3
|
|
edges.extend( [(face_vert_loc_indicies[i], face_vert_loc_indicies[i+1]) for i in xrange(len_face_vert_loc_indicies-1)] )
|
|
|
|
faces.pop(f_idx)
|
|
else:
|
|
|
|
# Smooth Group
|
|
if unique_smooth_groups and context_smooth_group and len_face_vert_loc_indicies > 3:
|
|
# Is a part of of a smooth group and is a face
|
|
if context_smooth_group_old is not context_smooth_group:
|
|
edge_dict= smooth_group_users[context_smooth_group]
|
|
context_smooth_group_old= context_smooth_group
|
|
|
|
for i in xrange(len_face_vert_loc_indicies):
|
|
i1= face_vert_loc_indicies[i]
|
|
i2= face_vert_loc_indicies[i-1]
|
|
if i1>i2: i1,i2= i2,i1
|
|
|
|
try:
|
|
edge_dict[i1,i2]+= 1
|
|
except KeyError:
|
|
edge_dict[i1,i2]= 1
|
|
|
|
# FGons into triangles
|
|
if has_ngons and len_face_vert_loc_indicies > 4:
|
|
|
|
ngon_face_indices= BPyMesh.ngon(verts_loc, face_vert_loc_indicies)
|
|
|
|
faces.extend(\
|
|
[(\
|
|
[face_vert_loc_indicies[ngon[0]], face_vert_loc_indicies[ngon[1]], face_vert_loc_indicies[ngon[2]] ],\
|
|
[face_vert_tex_indicies[ngon[0]], face_vert_tex_indicies[ngon[1]], face_vert_tex_indicies[ngon[2]] ],\
|
|
context_material,\
|
|
context_smooth_group,\
|
|
context_object)\
|
|
for ngon in ngon_face_indices]\
|
|
)
|
|
|
|
# edges to make fgons
|
|
if CREATE_FGONS:
|
|
edge_users= {}
|
|
for ngon in ngon_face_indices:
|
|
for i in (0,1,2):
|
|
i1= face_vert_loc_indicies[ngon[i ]]
|
|
i2= face_vert_loc_indicies[ngon[i-1]]
|
|
if i1>i2: i1,i2= i2,i1
|
|
|
|
try:
|
|
edge_users[i1,i2]+=1
|
|
except KeyError:
|
|
edge_users[i1,i2]= 1
|
|
|
|
for key, users in edge_users.iteritems():
|
|
if users>1:
|
|
fgon_edges[key]= None
|
|
|
|
# remove all after 3, means we dont have to pop this one.
|
|
faces.pop(f_idx)
|
|
|
|
|
|
# Build sharp edges
|
|
if unique_smooth_groups:
|
|
for edge_dict in smooth_group_users.itervalues():
|
|
for key, users in edge_dict.iteritems():
|
|
if users==1: # This edge is on the boundry of a group
|
|
sharp_edges[key]= None
|
|
|
|
|
|
# mat the material names to an index
|
|
material_mapping= dict([(name, i) for i, name in enumerate(unique_materials.keys())])
|
|
|
|
materials= [None] * len(unique_materials)
|
|
|
|
for name, index in material_mapping.iteritems():
|
|
materials[index]= unique_materials[name]
|
|
|
|
try:
|
|
me= Mesh.New(faces[0][5]) # object name from first face
|
|
except:
|
|
me= Mesh.New()
|
|
|
|
me.materials= materials[0:16] # make sure the list isnt too big.
|
|
#me.verts.extend([(0,0,0)]) # dummy vert
|
|
me.verts.extend(verts_loc)
|
|
|
|
face_mapping= me.faces.extend([f[0] for f in faces], indexList=True)
|
|
|
|
if verts_tex and me.faces:
|
|
me.faceUV= 1
|
|
# TEXMODE= Mesh.FaceModes['TEX']
|
|
|
|
context_material_old= -1 # avoid a dict lookup
|
|
mat= 0 # rare case it may be un-initialized.
|
|
me_faces= me.faces
|
|
ALPHA= Mesh.FaceTranspModes.ALPHA
|
|
|
|
for i, face in enumerate(faces):
|
|
if len(face[0])==2:
|
|
if CREATE_EDGES:
|
|
edges.append(face[0])
|
|
else:
|
|
face_index_map= face_mapping[i]
|
|
if face_index_map!=None: # None means the face wasnt added
|
|
blender_face= me.faces[face_index_map]
|
|
|
|
face_vert_loc_indicies,\
|
|
face_vert_tex_indicies,\
|
|
context_material,\
|
|
context_smooth_group,\
|
|
context_object= face
|
|
|
|
if context_smooth_group:
|
|
blender_face.smooth= True
|
|
|
|
if context_material:
|
|
if context_material_old is not context_material:
|
|
mat= material_mapping[context_material]
|
|
if mat>15:
|
|
mat= 15
|
|
context_material_old= context_material
|
|
|
|
blender_face.mat= mat
|
|
|
|
|
|
if verts_tex:
|
|
if context_material:
|
|
image= unique_material_images[context_material]
|
|
if image: # Can be none if the material dosnt have an image.
|
|
blender_face.image= image
|
|
if image.depth == 32:
|
|
blender_face.transp |= ALPHA
|
|
|
|
# BUG - Evil eekadoodle problem where faces that have vert index 0 location at 3 or 4 are shuffled.
|
|
if len(face_vert_loc_indicies)==4:
|
|
if face_vert_loc_indicies[2]==0 or face_vert_loc_indicies[3]==0:
|
|
face_vert_tex_indicies= face_vert_tex_indicies[2], face_vert_tex_indicies[3], face_vert_tex_indicies[0], face_vert_tex_indicies[1]
|
|
else: # length of 3
|
|
if face_vert_loc_indicies[2]==0:
|
|
face_vert_tex_indicies= face_vert_tex_indicies[1], face_vert_tex_indicies[2], face_vert_tex_indicies[0]
|
|
# END EEEKADOODLE FIX
|
|
|
|
# assign material, uv's and image
|
|
for ii, uv in enumerate(blender_face.uv):
|
|
uv.x, uv.y= verts_tex[face_vert_tex_indicies[ii]]
|
|
del me_faces
|
|
del ALPHA
|
|
|
|
# Add edge faces.
|
|
me_edges= me.edges
|
|
if CREATE_FGONS and fgon_edges:
|
|
FGON= Mesh.EdgeFlags.FGON
|
|
for ed in me.findEdges( fgon_edges.keys() ):
|
|
if ed!=None:
|
|
me_edges[ed].flag |= FGON
|
|
del FGON
|
|
|
|
if unique_smooth_groups and sharp_edges:
|
|
SHARP= Mesh.EdgeFlags.SHARP
|
|
for ed in me.findEdges( sharp_edges.keys() ):
|
|
if ed!=None:
|
|
me_edges[ed].flag |= SHARP
|
|
del SHARP
|
|
del me_edges
|
|
|
|
if CREATE_EDGES:
|
|
me.edges.extend( edges )
|
|
|
|
scn= Scene.GetCurrent()
|
|
ob= scn.objects.new(me)
|
|
new_objects.append(ob)
|
|
#ob= Object.New('Mesh')
|
|
#ob.link(me)
|
|
#scn.link(ob)
|
|
ob.sel= 1
|
|
# ob.makeDisplayList()
|
|
# me.update()
|
|
# print ob.getBoundBox()
|
|
|
|
|
|
|
|
def get_float_func(filepath):
|
|
'''
|
|
find the float function for this obj file
|
|
- weather to replace commas or not
|
|
'''
|
|
file= open(filepath, 'r')
|
|
for line in file.xreadlines():
|
|
if line.startswith('v'): # vn vt v
|
|
if ',' in line:
|
|
return lambda f: float(f.replace(',', '.'))
|
|
elif '.' in line:
|
|
return float
|
|
|
|
def load_obj(filepath, CLAMP_SIZE= 0.0, CREATE_FGONS= True, CREATE_SMOOTH_GROUPS= True, CREATE_EDGES= True, SPLIT_OBJECTS= True, SPLIT_GROUPS= True, SPLIT_MATERIALS= True, IMAGE_SEARCH=True):
|
|
'''
|
|
Called by the user interface or another script.
|
|
load_obj(path) - should give acceptable results.
|
|
This function passes the file and sends the data off
|
|
to be split into objects and then converted into mesh objects
|
|
'''
|
|
print '\nimporting obj "%s"' % filepath
|
|
|
|
time_main= sys.time()
|
|
|
|
verts_loc= []
|
|
verts_tex= []
|
|
faces= [] # tuples of the faces
|
|
material_libs= [] # filanems to material libs this uses
|
|
|
|
|
|
# Get the string to float conversion func for this file- is 'float' for almost all files.
|
|
float_func= get_float_func(filepath)
|
|
|
|
# Context variables
|
|
context_material= None
|
|
context_smooth_group= None
|
|
context_object= None
|
|
|
|
has_ngons= False
|
|
# has_smoothgroups= False - is explicit with len(unique_smooth_groups) being > 0
|
|
|
|
# Until we can use sets
|
|
unique_materials= {}
|
|
unique_material_images= {}
|
|
unique_smooth_groups= {}
|
|
# unique_obects= {} - no use for this variable since the objects are stored in the face.
|
|
|
|
# when there are faces that end with \
|
|
# it means they are multiline-
|
|
# since we use xreadline we cant skip to the next line
|
|
# so we need to know weather
|
|
multi_line_face= False
|
|
|
|
print '\tpassing obj file "%s"...' % filepath,
|
|
time_sub= sys.time()
|
|
file= open(filepath, 'r')
|
|
for line in file.xreadlines():
|
|
|
|
if line.startswith('v '):
|
|
line_split= line.split()
|
|
# rotate X90: (x,-z,y)
|
|
verts_loc.append( (float_func(line_split[1]), -float_func(line_split[3]), float_func(line_split[2])) )
|
|
|
|
elif line.startswith('vn '):
|
|
pass
|
|
|
|
elif line.startswith('vt '):
|
|
line_split= line.split()
|
|
verts_tex.append( (float_func(line_split[1]), float_func(line_split[2])) )
|
|
|
|
# Handel faces lines (as faces) and the second+ lines of fa multiline face here
|
|
# use 'f' not 'f ' because some objs (very rare have 'fo ' for faces)
|
|
elif line.startswith('f') or (line.startswith('l ') and CREATE_EDGES) or multi_line_face:
|
|
|
|
if multi_line_face:
|
|
# use face_vert_loc_indicies and face_vert_tex_indicies previously defined and used the obj_face
|
|
line_split= line.split()
|
|
multi_line_face= False
|
|
|
|
else:
|
|
line_split= line[2:].split()
|
|
face_vert_loc_indicies= []
|
|
face_vert_tex_indicies= []
|
|
|
|
# Instance a face
|
|
faces.append((\
|
|
face_vert_loc_indicies,\
|
|
face_vert_tex_indicies,\
|
|
context_material,\
|
|
context_smooth_group,\
|
|
context_object\
|
|
))
|
|
|
|
|
|
if line_split[-1][-1]== '\\':
|
|
multi_line_face= True
|
|
if len(line_split[-1])==1:
|
|
line_split.pop() # remove the \ item
|
|
else:
|
|
line_split[-1]= line_split[-1][:-1] # remove the \ from the end last number
|
|
|
|
isline= line.startswith('l')
|
|
|
|
for v in line_split:
|
|
obj_vert= v.split('/')
|
|
|
|
vert_loc_index= int(obj_vert[0])-1
|
|
|
|
# Make relative negative vert indicies absolute
|
|
if vert_loc_index < 0:
|
|
vert_loc_index= len(verts_loc) + vert_loc_index + 1
|
|
|
|
face_vert_loc_indicies.append(vert_loc_index)
|
|
|
|
if not isline:
|
|
if len(obj_vert)>1 and obj_vert[1]:
|
|
# formatting for faces with normals and textures us
|
|
# loc_index/tex_index/nor_index
|
|
|
|
vert_tex_index= int(obj_vert[1])-1
|
|
# Make relative negative vert indicies absolute
|
|
if vert_tex_index < 0:
|
|
vert_tex_index= len(verts_tex) + vert_tex_index + 1
|
|
|
|
face_vert_tex_indicies.append(vert_tex_index)
|
|
else:
|
|
# dummy
|
|
face_vert_tex_indicies.append(0)
|
|
|
|
if len(face_vert_loc_indicies) > 4:
|
|
has_ngons= True
|
|
|
|
elif line.startswith('s'):
|
|
if CREATE_SMOOTH_GROUPS:
|
|
context_smooth_group= line_value(line.split())
|
|
if context_smooth_group=='off':
|
|
context_smooth_group= None
|
|
elif context_smooth_group: # is not None
|
|
unique_smooth_groups[context_smooth_group]= None
|
|
|
|
elif line.startswith('o'):
|
|
if SPLIT_OBJECTS:
|
|
context_object= line_value(line.split())
|
|
# unique_obects[context_object]= None
|
|
|
|
elif line.startswith('g'):
|
|
if SPLIT_GROUPS:
|
|
context_object= line_value(line.split())
|
|
# print 'context_object', context_object
|
|
# unique_obects[context_object]= None
|
|
|
|
elif line.startswith('usemtl'):
|
|
context_material= line_value(line.split())
|
|
unique_materials[context_material]= None
|
|
elif line.startswith('mtllib'): # usemap or usemat
|
|
material_libs.extend( line.split()[1:] ) # can have multiple mtllib filenames per line
|
|
|
|
''' # How to use usemap? depricated?
|
|
elif line.startswith('usema'): # usemap or usemat
|
|
context_image= line_value(line.split())
|
|
'''
|
|
|
|
file.close()
|
|
time_new= sys.time()
|
|
print '%.4f sec' % (time_new-time_sub)
|
|
time_sub= time_new
|
|
|
|
|
|
print '\tloading materials and images...',
|
|
create_materials(filepath, material_libs, unique_materials, unique_material_images, IMAGE_SEARCH)
|
|
|
|
time_new= sys.time()
|
|
print '%.4f sec' % (time_new-time_sub)
|
|
time_sub= time_new
|
|
|
|
|
|
# deselect all
|
|
for ob in Scene.GetCurrent().objects: ob.sel= False
|
|
new_objects= [] # put new objects here
|
|
|
|
print '\tbuilding geometry;\n\tverts:%i faces:%i materials: %i smoothgroups:%i ...' % ( len(verts_loc), len(faces), len(unique_materials), len(unique_smooth_groups) ),
|
|
# Split the mesh by objects/materials, may
|
|
for verts_loc_split, faces_split, unique_materials_split in split_mesh(verts_loc, faces, unique_materials, SPLIT_OBJECTS, SPLIT_MATERIALS):
|
|
# Create meshes from the data
|
|
create_mesh(new_objects, has_ngons, CREATE_FGONS, CREATE_EDGES, verts_loc_split, verts_tex, faces_split, unique_materials_split, unique_material_images, unique_smooth_groups)
|
|
|
|
axis_min= [ 1000000000]*3
|
|
axis_max= [-1000000000]*3
|
|
|
|
if CLAMP_SIZE:
|
|
# Get all object bounds
|
|
for ob in new_objects:
|
|
for v in ob.getBoundBox():
|
|
for axis, value in enumerate(v):
|
|
if axis_min[axis] > value: axis_min[axis]= value
|
|
if axis_max[axis] < value: axis_max[axis]= value
|
|
|
|
# Scale objects
|
|
max_axis= max(axis_max[0]-axis_min[0], axis_max[1]-axis_min[1], axis_max[2]-axis_min[2])
|
|
scale= 1.0
|
|
|
|
while CLAMP_SIZE < max_axis * scale:
|
|
scale= scale/10.0
|
|
|
|
for ob in new_objects:
|
|
ob.setSize(scale, scale, scale)
|
|
|
|
time_new= sys.time()
|
|
|
|
print '%.4f sec' % (time_new-time_sub)
|
|
print 'finished importing: "%s" in %.4f sec.' % (filepath, (time_new-time_main))
|
|
|
|
|
|
DEBUG= True
|
|
|
|
|
|
def load_obj_ui(filepath, BATCH_LOAD= False):
|
|
if BPyMessages.Error_NoFile(filepath):
|
|
return
|
|
|
|
CREATE_SMOOTH_GROUPS= Draw.Create(0)
|
|
CREATE_FGONS= Draw.Create(1)
|
|
CREATE_EDGES= Draw.Create(1)
|
|
SPLIT_OBJECTS= Draw.Create(1)
|
|
SPLIT_GROUPS= Draw.Create(1)
|
|
SPLIT_MATERIALS= Draw.Create(1)
|
|
CLAMP_SIZE= Draw.Create(10.0)
|
|
IMAGE_SEARCH= Draw.Create(1)
|
|
|
|
|
|
# Get USER Options
|
|
pup_block= [\
|
|
('Smooth Groups', CREATE_SMOOTH_GROUPS, 'Surround smooth groups by sharp edges'),\
|
|
('Create FGons', CREATE_FGONS, 'Import faces with more then 4 verts as fgons.'),\
|
|
('Lines', CREATE_EDGES, 'Import lines and faces with 2 verts as edges'),\
|
|
('Split by Object', SPLIT_OBJECTS, 'Import OBJ Objects into Blender Objects'),\
|
|
('Split by Groups', SPLIT_GROUPS, 'Import OBJ Groups into Blender Objects'),\
|
|
('Split by Material', SPLIT_MATERIALS, 'Import each material into a seperate mesh (Avoids > 16 per mesh error)'),\
|
|
('Clamp Scale:', CLAMP_SIZE, 0.0, 1000.0, 'Clamp the size to this maximum (Zero to Disable)'),\
|
|
'',\
|
|
('Image Search', IMAGE_SEARCH, 'Search subdirs for any assosiated images (Warning, may be slow)'),\
|
|
]
|
|
|
|
|
|
if not Draw.PupBlock('Import OBJ...', pup_block):
|
|
return
|
|
|
|
Window.WaitCursor(1)
|
|
|
|
if BATCH_LOAD: # load the dir
|
|
try:
|
|
files= [ f for f in os.listdir(filepath) if f.lower().endswith('.obj') ]
|
|
except:
|
|
Draw.PupMenu('Error%t|Could not open path ' + filepath)
|
|
return
|
|
|
|
if not files:
|
|
Draw.PupMenu('Error%t|No files at path ' + filepath)
|
|
return
|
|
|
|
for f in files:
|
|
scn= Scene.New( stripExt(f) )
|
|
scn.makeCurrent()
|
|
|
|
load_obj(sys.join(filepath, f),\
|
|
CLAMP_SIZE.val,\
|
|
CREATE_FGONS.val,\
|
|
CREATE_SMOOTH_GROUPS.val,\
|
|
CREATE_EDGES.val,\
|
|
SPLIT_OBJECTS.val,\
|
|
SPLIT_GROUPS.val,\
|
|
SPLIT_MATERIALS.val,\
|
|
IMAGE_SEARCH.val,\
|
|
)
|
|
|
|
else: # Normal load
|
|
load_obj(filepath,\
|
|
CLAMP_SIZE.val,\
|
|
CREATE_FGONS.val,\
|
|
CREATE_SMOOTH_GROUPS.val,\
|
|
CREATE_EDGES.val,\
|
|
SPLIT_OBJECTS.val,\
|
|
SPLIT_GROUPS.val,\
|
|
SPLIT_MATERIALS.val,\
|
|
IMAGE_SEARCH.val,\
|
|
)
|
|
|
|
Window.WaitCursor(0)
|
|
|
|
|
|
def load_obj_ui_batch(file):
|
|
load_obj_ui(file, True)
|
|
|
|
DEBUG= False
|
|
|
|
if __name__=='__main__' and not DEBUG:
|
|
if os and Window.GetKeyQualifiers() & Window.Qual.SHIFT:
|
|
Window.FileSelector(load_obj_ui_batch, 'Import OBJ Dir', '')
|
|
else:
|
|
Window.FileSelector(load_obj_ui, 'Import a Wavefront OBJ', '*.obj')
|
|
|
|
|
|
'''
|
|
# For testing compatibility
|
|
else:
|
|
# DEBUG ONLY
|
|
TIME= sys.time()
|
|
import os
|
|
print 'Searching for files'
|
|
os.system('find /fe/obj -iname "*.obj" > /tmp/temp3ds_list')
|
|
|
|
print '...Done'
|
|
file= open('/tmp/temp3ds_list', 'r')
|
|
lines= file.readlines()
|
|
file.close()
|
|
|
|
def between(v,a,b):
|
|
if v <= max(a,b) and v >= min(a,b):
|
|
return True
|
|
return False
|
|
|
|
for i, _obj in enumerate(lines):
|
|
if between(i, 440,600):
|
|
_obj= _obj[:-1]
|
|
print 'Importing', _obj, '\nNUMBER', i, 'of', len(lines)
|
|
_obj_file= _obj.split('/')[-1].split('\\')[-1]
|
|
newScn= Scene.New(_obj_file)
|
|
newScn.makeCurrent()
|
|
load_obj(_obj, False)
|
|
|
|
print 'TOTAL TIME: %.6f' % (sys.time() - TIME)
|
|
'''
|
|
|
|
#load_obj('/test.obj')
|
|
#load_obj('/fe/obj/mba1.obj')
|