Skip to content

Instantly share code, notes, and snippets.

@shivshank
Last active December 31, 2023 01:12
Show Gist options
  • Star 43 You must be signed in to star a gist
  • Fork 9 You must be signed in to fork a gist
  • Save shivshank/34ffc56283273e1e904e8dc50491a7a6 to your computer and use it in GitHub Desktop.
Save shivshank/34ffc56283273e1e904e8dc50491a7a6 to your computer and use it in GitHub Desktop.
Exports from MagicaVoxel VOX to OBJ. Can preserve all edges for easy editing in a program like Blender.
"""
This script is designed to export a mass amount of MagicaVoxel .vox files
to .obj. Unlike Magica's internal exporter, this exporter preserves the
voxel vertices for easy manipulating in a 3d modeling program like Blender.
Various meshing algorithms are included (or to be included). MagicaVoxel
uses monotone triangulation (I think). The algorithms that will (or do)
appear in this script will use methods to potentially reduce rendering
artifacts that could be introduced by triangulation of this nature.
I may also include some features like light map generation for easy
importing into Unreal Engine, etc.
Notes:
* There may be a few floating point equality comparisons. They seem to
work but it scares me a little.
* TODO: use constants instead of magic numbers (as defined in AAQuad),
(i.e., ..., 2 -> AAQuad.TOP, ...)
* A lot of assertions should probably be exceptions since they are
error checking user input (this sounds really bad now that I've put
it on paper...). So don't run in optimized mode (who does that
anyways?).
* I am considering adding FBX support.
"""
import math
class AAQuad:
""" A solid colored axis aligned quad. """
normals = [
(-1, 0, 0), # left = 0
(1, 0, 0), # right = 1
(0, 0, 1), # top = 2
(0, 0, -1), # bottom = 3
(0, -1, 0), # front = 4
(0, 1, 0) # back = 5
]
LEFT = 0
RIGHT = 1
TOP = 2
BOTTOM = 3
FRONT = 4
BACK = 5
def __init__(self, verts, uv=None, normal=None):
assert len(verts) == 4, "face must be a quad"
self.vertices = verts
self.uv = uv
self.normal = normal
def __str__(self):
s = []
for i in self.vertices:
s.append( str(i) + '/' + str(self.uv) + '/' + str(self.normal))
return 'f ' + ' '.join(s)
def center(self):
return (
sum(i[0] for i in self.vertices)/4,
sum(i[1] for i in self.vertices)/4,
sum(i[2] for i in self.vertices)/4
)
def bucketHash(faces, origin, maximum, bucket=16):
extents = (
math.ceil((maximum[0] - origin[0])/bucket),
math.ceil((maximum[1] - origin[1])/bucket),
math.ceil((maximum[2] - origin[2])/bucket)
)
buckets = {}
for f in faces:
c = f.center()
# TODO
def optimizedGreedyMesh(faces):
# TODO
edges = adjacencyGraphEdges(faces)
groups = contiguousFaces(faces, edges)
return faces
def adjacencyGraphEdges(faces):
""" Get the list of edges representing adjacent faces. """
# a list of edges, where edges are tuple(face_a, face_b)
edges = []
# build the list of edges in the graph
for root in faces:
for face in faces:
if face is root:
continue
if facesAreAdjacent(root, face):
# the other edge will happen somewhere else in the iteration
# (i.e., the relation isAdjacent is symmetric)
edges.append((root, face))
return edges
def contiguousFaces(faces, adjacencyGraphEdges):
""" Get the list of connected components from a list of graph edges.
The list will contain lists containing the edges within the components.
"""
groups = []
visited = dict((f, False) for f in faces)
for face in faces:
# if the face hasn't been visited, it is not in any found components
if not visited[face]:
g = []
_visitGraphNodes(face, adjacencyGraphEdges, visited, g)
# there is only a new component if face has not been visited yet
groups.append(g)
return groups
def _visitGraphNodes(node, edges, visited, component):
""" Recursive routine used in findGraphComponents """
# visit every component connected to this one
for edge in edges:
# for all x in nodes, (node, x) and (x, node) should be in edges!
# therefore we don't have to check for "edge[1] is node"
if edge[0] is node and not visited[edge[1]]:
assert edge[1] is not node, "(node, node) should not be in edges"
# mark the other node as visited
visited[edge[1]] = True
component.append(edge[1])
# visit all of that nodes connected nodes
_visitGraphNodes(edge[1], edges, visited, component)
def facesAreAdjacent(a, b):
""" Adjacent is defined as same normal, uv, and a shared edge.
This isn't entirely intuitive (i.e., corner faces are not adjacent)
but this definition fits the problem domain.
Only works on AAQuads.
"""
# note: None is == None, this shouldn't matter
if a.uv != b.uv:
return False
if a.normal != b.normal:
return False
# to be adjacent, two faces must share an edge
# use == and not identity in case edge split was used
shared = 0
for vert_a in a.vertices:
for vert_b in b.vertices:
if vert_a == vert_b:
shared += 1
# hooray we have found a shared edge (or a degenerate case...)
if shared == 2:
return True
return False
class GeoFace:
""" An arbitrary geometry face
This should only be used for arbitrary models, not ones we can
reasonably assume are axis aligned.
"""
def __init__(self, verts, uvs=None, normals=None):
self.vertices = verts
assert len(verts) in (3, 4), "only quads and tris are supported"
self.normals = normals
self.uvs = uvs
def toAAQuad(self, skipAssert=False):
q = AAQuad(self.vertices)
if self.normals is not None and len(self.normals) > 0:
if not skipAssert:
for i in self.normals:
assert self.normals[0] == i, \
"face must be axis aligned (orthogonal normals)"
q.normal = self.normals[0]
if self.uvs is not None and len(self.uvs) > 0:
if not skipAssert:
for i in self.uvs:
assert self.uvs[0] == i, \
"face must be axis aligned (orthogonal)"
q.uv = self.uvs[0]
return q
class VoxelStruct:
""" Describes a voxel object
"""
def __init__(self):
# a dict is probably the best way to go about this
# (as a trade off between performance and code complexity)
# see _index for the indexing method
self.voxels = {}
self.colorIndices = set()
def fromList(self, voxels):
self.voxels = {}
for voxel in voxels:
self.setVoxel(voxel)
self.colorIndices.add(voxel.colorIndex)
def setVoxel(self, voxel):
self.voxels[voxel.z*(256**2) + voxel.y * 256 + voxel.x] = voxel
def getVoxel(self, x, y, z):
return self.voxels.get(z*(256**2) + y * 256 + x, None)
def _index(self, x, y, z):
return z*(256**2) + y * 256 + x
def getBounds(self):
origin = (float("inf"), float("inf"), float("inf"))
maximum = (float("-inf"), float("-inf"), float("-inf"))
for key, voxel in self.voxels.items():
origin = (
min(origin[0], voxel.x),
min(origin[1], voxel.y),
min(origin[2], voxel.z)
)
maximum = (
max(maximum[0], voxel.x),
max(maximum[1], voxel.y),
max(maximum[2], voxel.z)
)
return origin, maximum
def zeroOrigin(self):
""" Translate the model so that it's origin is at 0, 0, 0 """
origin, maximum = self.getBounds()
result = {}
xOff, yOff, zOff = origin
for key, voxel in self.voxels.iteritems():
result[self._index(voxel.x-xOff, voxel.y-yOff, voxel.z-zOff)] = \
Voxel(voxel.x-xOff, voxel.y-yOff, voxel.z-zOff,
voxel.colorIndex)
self.voxels = result
return (0, 0, 0), (maximum[0] - xOff,
maximum[1] - yOff,
maximum[2] - zOff)
def toQuads(self):
""" --> a list of AAQuads """
faces = []
for key, voxel in self.voxels.items():
self._getObjFaces(voxel, faces)
return faces
def _getObjFaces(self, voxel, outFaces):
if voxel.colorIndex == 0:
# do nothing if this is an empty voxel
# n.b., I do not know if this ever can happen.
return []
sides = self._objExposed(voxel)
if sides[0]:
f = self._getLeftSide(voxel)
self._getObjFacesSupport(0, voxel.colorIndex, f, outFaces)
if sides[1]:
f = self._getRightSide(voxel)
self._getObjFacesSupport(1, voxel.colorIndex, f, outFaces)
if sides[2]:
f = self._getTopSide(voxel)
self._getObjFacesSupport(2, voxel.colorIndex, f, outFaces)
if sides[3]:
f = self._getBottomSide(voxel)
self._getObjFacesSupport(3, voxel.colorIndex, f, outFaces)
if sides[4]:
f = self._getFrontSide(voxel)
self._getObjFacesSupport(4, voxel.colorIndex, f, outFaces)
if sides[5]:
f = self._getBackSide(voxel)
self._getObjFacesSupport(5, voxel.colorIndex, f, outFaces)
return
n = AAQuad.normals[i]
# note: texcoords are based on MagicaVoxel's texturing scheme!
# meaning a color index of 0 translates to pixel[255]
# and color index [1:256] -> pixel[0:255]
u = ((voxel.colorIndex - 1)/256 + 1/512, 0.5)
outFaces.append(
# this is most definitely not "fun"
AAQuad(f, u, n)
)
def _getObjFacesSupport(self, side, color, faces, outFaces):
n = AAQuad.normals[side]
# note: texcoords are based on MagicaVoxel's texturing scheme!
# meaning a color index of 0 translates to pixel[255]
# and color index [1:256] -> pixel[0:255]
u = ((color - 1)/256 + 1/512, 0.5)
outFaces.append(
# fact: the parameters were coincidentally "f, u, n" at one point!
AAQuad(faces, u, n)
)
# MagicaVoxel does -.5 to +.5 for each cube, we'll do 0.0 to 1.0 ;)
def _getLeftSide(self, voxel):
return [
(voxel.x, voxel.y + 1, voxel.z + 1),
(voxel.x, voxel.y + 1, voxel.z),
(voxel.x, voxel.y, voxel.z),
(voxel.x, voxel.y, voxel.z + 1)
]
def _getRightSide(self, voxel):
return (
(voxel.x + 1, voxel.y, voxel.z + 1),
(voxel.x + 1, voxel.y, voxel.z),
(voxel.x + 1, voxel.y + 1, voxel.z),
(voxel.x + 1, voxel.y + 1, voxel.z + 1)
)
def _getTopSide(self, voxel):
return (
(voxel.x, voxel.y + 1, voxel.z + 1),
(voxel.x, voxel.y, voxel.z + 1),
(voxel.x + 1, voxel.y, voxel.z + 1),
(voxel.x + 1, voxel.y + 1, voxel.z + 1)
)
def _getBottomSide(self, voxel):
return (
(voxel.x, voxel.y, voxel.z),
(voxel.x, voxel.y + 1, voxel.z),
(voxel.x + 1, voxel.y + 1, voxel.z),
(voxel.x + 1, voxel.y, voxel.z)
)
def _getFrontSide(self, voxel):
return (
(voxel.x, voxel.y, voxel.z + 1),
(voxel.x, voxel.y, voxel.z),
(voxel.x + 1, voxel.y, voxel.z),
(voxel.x + 1, voxel.y, voxel.z + 1)
)
def _getBackSide(self, voxel):
return (
(voxel.x + 1, voxel.y + 1, voxel.z + 1),
(voxel.x + 1, voxel.y + 1, voxel.z),
(voxel.x, voxel.y + 1, voxel.z),
(voxel.x, voxel.y + 1, voxel.z + 1)
)
def _objExposed(self, voxel):
""" --> a set of [0, 6) representing which voxel faces are shown
for the meaning of 0-5, see AAQuad.normals
get the sick truth about these voxels' dirty secrets...
"""
# check left 0
side = self.getVoxel(voxel.x - 1, voxel.y, voxel.z)
s0 = side is None or side.colorIndex == 0
# check right 1
side = self.getVoxel(voxel.x + 1, voxel.y, voxel.z)
s1 = side is None or side.colorIndex == 0
# check top 2
side = self.getVoxel(voxel.x, voxel.y, voxel.z + 1)
s2 = side is None or side.colorIndex == 0
# check bottom 3
side = self.getVoxel(voxel.x, voxel.y, voxel.z - 1)
s3 = side is None or side.colorIndex == 0
# check front 4
side = self.getVoxel(voxel.x, voxel.y - 1, voxel.z)
s4 = side is None or side.colorIndex == 0
# check back 5
side = self.getVoxel(voxel.x, voxel.y + 1, voxel.z)
s5 = side is None or side.colorIndex == 0
return s0, s1, s2, s3, s4, s5
class Voxel:
def __init__(self, x, y, z, colorIndex):
self.x = x
self.y = y
self.z = z
self.colorIndex = colorIndex
def genNormals(self, aaQuads, overwrite=False):
# compute CCW normal if it doesn't exist
for face in aaQuads:
if overwrite or face.normal is None:
side_a = (face.vertices[1][0] - face.vertices[0][0],
face.vertices[1][1] - face.vertices[0][1],
face.vertices[1][2] - face.vertices[0][2])
side_b = (face.vertices[-1][0] - face.vertices[0][0],
face.vertices[-1][1] - face.vertices[0][1],
face.vertices[-1][2] - face.vertices[0][2])
# compute the cross product
face.normal = (side_a[1]*side_b[2] - side_a[2]*side_b[1],
side_a[2]*side_b[0] - side_a[0]*side_b[2],
side_a[0]*side_b[1] - side_a[1]*side_b[0])
def importObj(stream):
vertices = []
faces = []
uvs = []
normals = []
for line in stream:
# make sure there's no new line or trailing spaces
l = line.strip().split(' ')
lineType = l[0].strip()
data = l[1:]
if lineType == 'v':
# vertex
v = tuple(map(float, data))
vertices.append(v)
elif lineType == 'vt':
# uv
uvs.append( tuple(map(float, data)) )
elif lineType == 'vn':
# normal
normals.append( tuple(map(float, data)) )
elif lineType == 'f':
# face (assume all verts/uvs/normals have been processed)
faceVerts = []
faceUvs = []
faceNormals = []
for v in data:
result = v.split('/')
print(result)
# recall that everything is 1 indexed...
faceVerts.append(vertices[int(result[0]) - 1])
if len(result) == 1:
continue # there is only a vertex index
if result[1] != '':
# uvs may not be present, ex: 'f vert//normal ...'
faceUvs.append(uvs[int(result[1]) - 1])
if len(result) <= 2:
# don't continue if only vert and uv are present
continue
faceNormals.append(normals[int(result[2]) - 1])
faces.append( GeoFace(faceVerts, faceUvs, faceNormals) )
else:
# there could be material specs, smoothing, or comments... ignore!
pass
return faces
def exportObj(stream, aaQuads):
# gather some of the needed information
faces = aaQuads
# copy the normals from AAQuad (99% of cases will use all directions)
normals = list(AAQuad.normals)
uvs = set()
for f in faces:
if f.uv is not None:
uvs.add(f.uv)
# convert this to a list because we need to get their index later
uvs = list(uvs)
# we will build a list of vertices as we go and then write everything
# in bulk, disadvantage that MANY verts will be duplicated in the OBJ file
fLines = []
vertices = []
indexOffset = 0
for f in faces:
# recall that OBJ files are 1 indexed
n = 1 + normals.index(f.normal) if f.normal is not None else ''
uv = 1 + uvs.index(f.uv) if f.uv is not None else ''
# this used to be a one liner ;)
fLine = ['f']
for i, vert in enumerate(f.vertices):
# for each vertex of this face
v = 1 + indexOffset + f.vertices.index(vert)
fLine.append(str(v) + '/' + str(uv) + '/' + str(n))
vertices.extend(f.vertices)
indexOffset += len(f.vertices)
fLines.append(' '.join(fLine) + '\n')
# write to the file
stream.write('# shivshank\'s .obj optimizer\n')
stream.write('\n')
if len(normals) > 0:
stream.write('# normals\n')
for n in normals:
stream.write('vn ' + ' '.join(list(map(str, n))) + '\n')
stream.write('\n')
if len(uvs) > 0:
stream.write('# texcoords\n')
for i in uvs:
stream.write('vt ' + ' '.join(list(map(str, i))) + '\n')
stream.write('\n')
# output the vertices and faces
stream.write('# verts\n')
for v in vertices:
stream.write('v ' + ' '.join(list(map(str, v))) + '\n')
stream.write('\n')
stream.write('# faces\n')
for i in fLines:
stream.write(i)
stream.write('\n')
stream.write('\n')
return len(vertices), len(fLines)
def importVox(file):
""" --> a VoxelStruct from this .vox file stream """
# in theory this could elegantly be many functions and classes
# but this is such a simple file format...
# refactor: ? should probably find a better exception type than value error
vox = VoxelStruct()
magic = file.read(4)
if magic != b'VOX ':
print('magic number is', magic)
if userAborts('This does not appear to be a VOX file. Abort?'):
raise ValueError("Invalid magic number")
# the file appears to use little endian consistent with RIFF
version = int.from_bytes(file.read(4), byteorder='little')
if version != 150:
if userAborts('Only version 150 is supported; this file: '
+ str(version) + '. Abort?'):
raise ValueError("Invalid file version")
mainHeader = _readChunkHeader(file)
if mainHeader['id'] != b'MAIN':
print('chunk id:', mainId)
if userAborts('Did not find the main chunk. Abort?'):
raise ValueError("Did not find main VOX chunk. ")
#assert mainHeader['size'] == 0, "main chunk should have size 0"
# we don't need anything from the size or palette header!
# : we can figure out (minimum) bounds later from the voxel data
# : we only need UVs from voxel data; user can export palette elsewhere
nextHeader = _readChunkHeader(file)
while nextHeader['id'] != b'XYZI':
# skip the contents of this header and its children, read the next one
file.read(nextHeader['size'] + nextHeader['childrenSize'])
nextHeader = _readChunkHeader(file)
voxelHeader = nextHeader
assert voxelHeader['id'] == b'XYZI', 'this should be literally impossible'
assert voxelHeader['childrenSize'] == 0, 'why voxel chunk have children?'
seekPos = file.tell()
totalVoxels = int.from_bytes(file.read(4), byteorder='little')
### READ THE VOXELS ###
for i in range(totalVoxels):
# n.b., byte order should be irrelevant since these are all 1 byte
x = int.from_bytes(file.read(1), byteorder='little')
y = int.from_bytes(file.read(1), byteorder='little')
z = int.from_bytes(file.read(1), byteorder='little')
color = int.from_bytes(file.read(1), byteorder='little')
vox.setVoxel(Voxel(x, y, z, color))
# assert that we've read the entire voxel chunk
assert file.tell() - seekPos == voxelHeader['size']
# (there may be more chunks after this but we don't need them!)
#print('\tdone reading voxel data;', totalVoxels , 'voxels read ;D')
return vox
def _readChunkHeader(buffer):
id = buffer.read(4)
if id == b'':
raise ValueError("Unexpected EOF, expected chunk header")
size = int.from_bytes(buffer.read(4), byteorder='little')
childrenSize = int.from_bytes(buffer.read(4), byteorder='little')
return {
'id': id, 'size': size, 'childrenSize': childrenSize
}
def userAborts(msg):
print(msg + ' (y/n)')
u = input()
if u.startswith('n'):
return False
return True
def exportAll():
""" Uses a file to automatically export a bunch of files!
See this function for details on the what the file looks like.
"""
import os, os.path
with open('exporter.txt', mode='r') as file:
# use this as a file "spec"
fromSource = os.path.abspath(file.readline().strip())
toExportDir = os.path.abspath(file.readline().strip())
optimizing = file.readline()
if optimizing.lower() == 'true':
optimizing = True
else:
optimizing = False
print('exporting vox files under', fromSource)
print('\tto directory', toExportDir)
print('\toptimizing?', optimizing)
print()
# export EVERYTHING (.vox) walking the directory structure
for p, dirList, fileList in os.walk(fromSource):
pathDiff = os.path.relpath(p, start=fromSource)
outDir = os.path.join(toExportDir, pathDiff)
# REFACTOR: the loop should be moved to a function
for fileName in fileList:
# only take vox files
if os.path.splitext(fileName)[1] != '.vox':
print('ignored', fileName)
continue
print('exporting', fileName)
# read/import the voxel file
with open(os.path.join(p, fileName), mode='rb') as file:
try:
vox = importVox(file)
except ValueError as exc:
print('aborted', fileName, str(exc))
continue
# mirror the directory structure in the export folder
if not os.path.exists(outDir):
os.makedirs(outDir)
print('\tcreated directory', outDir)
# export a non-optimized version
objName = os.path.splitext(fileName)[0]
rawQuads = vox.toQuads()
with open(os.path.join(outDir, objName + '.obj'), mode='w') as file:
vCount, qCount = exportObj(file, rawQuads)
print('\texported', vCount, 'vertices,', qCount, 'quads')
if optimizing:
# TODO
continue
optiFaces = optimizedGreedyMesh(rawQuads)
bucketHash(optiFaces, *vox.getBounds())
with open(os.path.join(outDir, objName + '.greedy.obj'),
mode='w') as file:
exportObj(file, optiFaces)
def byPrompt():
import os, os.path
from glob import glob
print('Enter an output path:')
u = input('> ').strip()
while not os.path.exists(u):
print('That path does not exist.')
print('Enter an output path:')
u = input('> ').strip()
outRoot = os.path.abspath(u)
print('Are we optimizing? (y/n)')
u = input('> ').strip()
# this could be a one liner but I think it's easier to read this way
if u.startswith('y'):
optimizing = True
else:
optimizing = False
try:
while True:
print('Enter glob of export files (\'exit\' or blank to quit):')
u = input('> ').strip()
if u == 'exit' or u == '':
break
u = glob(u)
for f in u:
print('reading VOX file', f)
with open(f, mode='rb') as file:
try:
vox = importVox(file)
except ValueError:
print('\tfile reading aborted')
continue
outFile = os.path.splitext(os.path.basename(f))[0]
outPath = os.path.join(outRoot, outFile)
print('exporting VOX to OBJ at path', outPath)
with open(outPath, mode='w') as file:
exportObj(file, vox.toQuads())
if optimizing:
# TODO
pass
except KeyboardInterrupt:
pass
if __name__ == "__main__":
profiling = False
try:
import cProfile
if profiling:
cProfile.run('exportAll()', sort='tottime')
else:
exportAll()
except OSError:
print('No instruction file found, falling back to prompt.')
byPrompt()
@Claytone
Copy link

Claytone commented Nov 15, 2021

This code is now available as a Python package! I've added some features including support for materials.

  • I will be adding conversion of entire folders.
  • I am definitely going to refine the export algorithm as it is O(n^2) and veeery slow with extremely large models.
  • I may add multi-object support at some point since multiple people have asked.
  • If you want to contribute, the repo is here https://gitlab.com/Claytone/vox2obj
  • If you just want to check it out, here's the PyPi link https://pypi.org/project/vox2obj/

For now, you can use vox2obj by installing it with pip and then

from vox2obj import convert_to_obj
convert_to_obj("test.vox", "test_converted.obj", "test.png")

The example code I wrote in previous messages of this thread should still work... but you should really use the pip package for the most recent version.

@SGNoaidi
Copy link

does this work on linux?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment