Skip to content

Instantly share code, notes, and snippets.

@smealum
Last active July 10, 2023 05:50
Show Gist options
  • Save smealum/8807124 to your computer and use it in GitHub Desktop.
Save smealum/8807124 to your computer and use it in GitHub Desktop.
import struct
import math
import os
import sys
from PIL import Image
from subprocess import call
if sys.version > '3':
buffer = memoryview
def getWord(b, k, n=4):
return sum(list(map(lambda c: b[k+c]<<(c*8),range(n))))
totalVcount=1
currentVcount=0
totalNcount=1
currentNcount=0
hasNormals=False
totalTCcount=1
currentTCcount=0
def parseVertices(b, o, s, f):
global hasNormals
global currentVcount, currentNcount, currentTCcount
if f==0x0C:
#for 3-word vertex descriptors
for i in range(int(s/f)):
arr=[struct.unpack('f',buffer(srcdata[(o+k*4):(o+(k+1)*4)]))[0] for k in range(3)]
print("v "+str(arr[0])+" "+str(arr[1])+" "+str(arr[2]))
currentVcount=currentVcount+1
o=o+f
if f==0x14:
#for 5-word vertex descriptors
for i in range(int(s/f)):
arr=[struct.unpack('f',buffer(srcdata[(o+k*4):(o+(k+1)*4)]))[0] for k in range(5)]
print("v "+str(arr[0])+" "+str(arr[1])+" "+str(arr[2]))
print("vt "+str(arr[3])+" "+str(arr[4]))
currentVcount=currentVcount+1
currentTCcount=currentTCcount+1
o=o+f
elif f==0x18:
#for 6-word vertex descriptors
for i in range(int(s/f)):
arr=[struct.unpack('f',buffer(srcdata[(o+k*4):(o+(k+1)*4)]))[0] for k in range(6)]
print("v "+str(arr[0])+" "+str(arr[1])+" "+str(arr[2]))
print("vt "+str(arr[4])+" "+str(arr[5]))
currentVcount=currentVcount+1
currentTCcount=currentTCcount+1
o=o+f
elif f==0x20 or f==0x24 or f==0x28:
#for 8-word vertex descriptors
#for 9-word vertex descriptors
#for 10-word vertex descriptors
for i in range(int(s/f)):
arr=[struct.unpack('f',buffer(srcdata[(o+k*4):(o+(k+1)*4)]))[0] for k in range(8)]
print("v "+str(arr[0])+" "+str(arr[1])+" "+str(arr[2]))
print("vn "+str(arr[3])+" "+str(arr[4])+" "+str(arr[5]))
print("vt "+str(arr[6])+" "+str(arr[7]))
hasNormals=True
currentVcount=currentVcount+1
currentNcount=currentNcount+1
currentTCcount=currentTCcount+1
o=o+f
def parseFaces(b, o, s, f):
global hasNormals
global totalVcount, totalNcount, totalTCcount
global currentVcount, currentNcount, currentTCcount
if f==1:
#for u8 face descriptors
for i in range(int(s/(3))):
arr=[struct.unpack('B',buffer(srcdata[(o+k):(o+(k+1))]))[0] for k in range(8)]
if hasNormals:
print("f "+str(arr[0]+totalVcount)+"/"+str(arr[0]+totalTCcount)+"/"+str(arr[0]+totalNcount)+" "+
str(arr[1]+totalVcount)+"/"+str(arr[1]+totalTCcount)+"/"+str(arr[1]+totalNcount)+" "+
str(arr[2]+totalVcount)+"/"+str(arr[2]+totalTCcount)+"/"+str(arr[2]+totalNcount))
else:
print("f "+str(arr[0]+totalVcount)+" "+str(arr[1]+totalVcount)+" "+str(arr[2]+totalVcount))
o=o+3
elif f==3:
#for u16 face descriptors
for i in range(int(s/(3*2))):
arr=[struct.unpack('H',buffer(srcdata[(o+k*2):(o+(k+1)*2)]))[0] for k in range(8)]
if hasNormals:
print("f "+str(arr[0]+totalVcount)+"/"+str(arr[0]+totalTCcount)+"/"+str(arr[0]+totalNcount)+" "+
str(arr[1]+totalVcount)+"/"+str(arr[1]+totalTCcount)+"/"+str(arr[1]+totalNcount)+" "+
str(arr[2]+totalVcount)+"/"+str(arr[2]+totalTCcount)+"/"+str(arr[2]+totalNcount))
else:
print("f "+str(arr[0]+totalVcount)+" "+str(arr[1]+totalVcount)+" "+str(arr[2]+totalVcount))
o=o+6
totalVcount=totalVcount+currentVcount
totalNcount=totalNcount+currentNcount
totalTCcount=totalTCcount+currentTCcount
currentVcount=0
currentNcount=0
currentTCcount=0
def parseSOBJ(b, o):
global hasNormals
dataStructOffset=getWord(b, o+0x44)
print("# SOBJ "+hex(o))
hasNormals=False
if(dataStructOffset!=0):
dataStructOffset=o+dataStructOffset
faceDataFormat=getWord(b,dataStructOffset+0x18,1)
faceDataSize=getWord(b,dataStructOffset+0x20)
faceDataOffset=dataStructOffset+0x24+getWord(b,dataStructOffset+0x24)
vertexDataFormat=getWord(b,dataStructOffset+0x68,1)
vertexDataSize=getWord(b,dataStructOffset+0x58)
vertexDataOffset=dataStructOffset+0x5C+getWord(b,dataStructOffset+0x5C)
print("# vertex format "+hex(vertexDataFormat))
parseVertices(b, vertexDataOffset, vertexDataSize, vertexDataFormat)
print("g SOBJ_"+hex(o))
parseFaces(b, faceDataOffset, faceDataSize, faceDataFormat)
#returns RGBA tuple (with added pixel size)
#format ID only confirmed for 0, 4, 5 and 7 !
def parsePixel(b, o, f):
if f==0:
#RGBA8
pixel=getWord(b, o, 1)
return ((getWord(b, o+3, 1), getWord(b, o+2, 1), getWord(b, o+1, 1), getWord(b, o, 1)), 4)
elif f==1:
#RGB8
pixel=getWord(b, o, 1)
return ((getWord(b, o, 1), getWord(b, o+1, 1), getWord(b, o+2, 1), 255), 3)
elif f==2:
#RGBA5551
pixel=getWord(b, o, 2)
return ((((pixel>>11)&0x1F)*8, ((pixel>>6)&0x1F)*8, ((pixel>>1)&0x1F)*8, (255 if pixel&0x1==1 else 0)), 2)
elif f==3:
#RGB565
pixel=getWord(b, o, 2)
return ((((pixel>>11)&0x1F)*8, ((pixel>>5)&0x3F)*4, ((pixel)&0x1F)*8, 255), 2)
elif f==4:
#RGBA4
pixel=getWord(b, o, 2)
return ((((pixel>>12)&0xF)*16, ((pixel>>8)&0xF)*16, ((pixel>>4)&0xF)*16, (pixel&0xF)*16), 2)
elif f==5:
#LA8
pixel=getWord(b, o+1, 1)
return ((pixel, pixel, pixel, getWord(b, o, 1)),2)
elif f==7:
#L8
pixel=getWord(b, o, 1)
return ((pixel, pixel, pixel, 255),1)
# elif f==0xA:
# #?
# return ((0,0,0,0), 1)
# elif f==0xC:
# #?
# return ((0,0,0,0), 1)
elif f==0xD:
#A4 ?
return ((0,0,0,0), 1)
else:
print("# unknown format : "+hex(f))
return ((0,0,0,0), 1)
tileOrder=[0,1,8,9,2,3,10,11,16,17,24,25,18,19,26,27,4,5,12,13,6,7,14,15,20,21,28,29,22,23,30,31,32,33,40,41,34,35,42,43,48,49,56,57,50,51,58,59,36,37,44,45,38,39,46,47,52,53,60,61,54,55,62,63]
def parseTile(im, b, o, f, x, y):
global tileOrder
for k in range(8*8):
i=tileOrder[k]%8
j=int((tileOrder[k]-i)/8)
pixel=parsePixel(b,o,f)
im.putpixel((x+i,y+j), pixel[0])
o=o+pixel[1]
return o
def parseTexture(b, o, s, w, h, f):
im = Image.new("RGB", (w,h))
dstname="tex_"+hex(o)+".png"
if f==0xB or f==0xC:
#ETC1; used external etc program (see https://gist.github.com/smealum/8897237 )
open("tmp_etc","wb").write(b[o:(o+s)])
#not super clean
call(["etc.exe", "tmp_etc", str(w), str(h)])
imgData=bytearray(open("tmp_etc.data","rb").read())
for j in range(0,h):
for i in range(0,w):
k=(i+j*w)*4
im.putpixel((i,j), (imgData[k],imgData[k+1],imgData[k+2],imgData[k+3]))
else:
for j in range(0,h,8):
for i in range(0,w,8):
o=parseTile(im, b, o, f, i, j)
im.save(dstname)
def parseTXOB(b, o):
textureDataWidth=getWord(b,o+0x18)
textureDataHeight=getWord(b,o+0x14)
if textureDataWidth!=0 and textureDataHeight!=0:
textureDataFormat=getWord(b,o+0x30,1)
textureDataSize=getWord(b,o+0x40)
textureDataOffset=o+0x44+getWord(b,o+0x44)
#o=o+0x20
#print(hex(textureDataOffset)+" : "+hex(getWord(b,o))+" "+hex(getWord(b,o+4*1))+" "+hex(getWord(b,o+4*2))+" "+hex(getWord(b,o+4*3))+" "+hex(getWord(b,o+4*4))+" "+hex(getWord(b,o+4*5))+" "+hex(getWord(b,o+4*6))+" "+hex(getWord(b,o+4*7))+" "+
# hex(getWord(b,o+4*8))+" "+hex(getWord(b,o+4*9))+" "+hex(getWord(b,o+4*10))+" ")
parseTexture(b, textureDataOffset, textureDataSize, textureDataWidth, textureDataHeight, textureDataFormat)
srcfn=sys.argv[1]
srcdata=bytearray(open(srcfn, 'rb').read())
for i in range(int(len(srcdata)/4)):
#scan for SOBJ
if getWord(srcdata, i*4)==0x4A424F53:
parseSOBJ(srcdata,i*4)
# #scan for TXOB
if getWord(srcdata, i*4)==0x424F5854:
parseTXOB(srcdata,i*4)
@jabberwockxeno
Copy link

So, how do we actually make use of that script?

@pakoito
Copy link

pakoito commented Feb 9, 2014

Some cool stuff you have here. Where did you find the cgfx file format specification?

@smealum
Copy link
Author

smealum commented Feb 11, 2014

@jabberwockxeno : just run it on a cgfx file.

@pakoito : i didn't find it anywhere, I REd it.

@jabberwockxeno
Copy link

@smealum: I'm assuming that's the native format that the models are in? Obviously, that requires us to extract them. How would we do that?

@gervoson
Copy link

gervoson commented Dec 1, 2014

how to use it ??

@vPumpking
Copy link

@smealum This script generates broken obj files. Can you fix it or give something to repair the files once done?

@pakoito
Copy link

pakoito commented Jul 9, 2023

True necroposting here

@vPumpking
Copy link

yeah 😆
but I have no choice
I need a commandline tool to extract cgfx files

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