shoebox导出的散图各帧的中心点不统一,这种情况可以通过该图集对应的json或者plist文件读取单张图片的区域进行导出。
##通常可使用shoebox
官网地址:http://renderhjs.net/shoebox/
## json格式
```
# -*- coding: utf-8 -*-
import os,sys
import json
import os
import os.path
from PIL import Image
def json_to_dict(json_filename):
json_file = open(json_filename, 'r')
all_pic_dic = json.load(json_file)
all_item_list = []
for one_pic_item in all_pic_dic['res']:
one_json_item = all_pic_dic['res'][one_pic_item]
one_item = {}
one_item['name'] = one_pic_item.strip().lstrip().rstrip(',')
one_item['x'] = one_json_item['x']
one_item['y'] = one_json_item['y']
one_item['w'] = one_json_item['w']
one_item['h'] = one_json_item['h']
all_item_list.append(one_item)
return all_item_list
def gen_png_from_json(folder_name, json_filename, png_filename):
big_image = Image.open(png_filename)
all_item_list = json_to_dict(json_filename)
print 'gen_png_from_json:' + folder_name
#清理掉原目录
#os.removedirs(folder_name)
os.mkdir(folder_name)
for i, one_item_data in enumerate(all_item_list):
file_name = one_item_data['name']
x = one_item_data['x']
y = one_item_data['y']
w = one_item_data['w']
h = one_item_data['h']
#设置图像裁剪区域 (x左上,y左上,x右下,y右下)
image_box = [x, y, x + w , y + h ]
one_pic = big_image.crop(image_box)
one_pic.save(folder_name + "/" + file_name + '.png') # 存储裁剪得到的图像
#print one_item_data
```
# plist格式
```
#!python
import os,sys
from xml.etree import ElementTree
from PIL import Image
def tree_to_dict(tree):
d = {}
for index, item in enumerate(tree):
if item.tag == 'key':
if tree[index+1].tag == 'string':
d[item.text] = tree[index + 1].text
elif tree[index + 1].tag == 'true':
d[item.text] = True
elif tree[index + 1].tag == 'false':
d[item.text] = False
elif tree[index+1].tag == 'dict':
d[item.text] = tree_to_dict(tree[index+1])
return d
def gen_png_from_plist(plist_filename, png_filename):
file_path = plist_filename.replace('.plist', '')
big_image = Image.open(png_filename)
root = ElementTree.fromstring(open(plist_filename, 'r').read())
plist_dict = tree_to_dict(root[0])
to_list = lambda x: x.replace('{','').replace('}','').split(',')
for k,v in plist_dict['frames'].items():
rectlist = to_list(v['frame'])
width = int( rectlist[3] if v['rotated'] else rectlist[2] )
height = int( rectlist[2] if v['rotated'] else rectlist[3] )
box=(
int(rectlist[0]),
int(rectlist[1]),
int(rectlist[0]) + width,
int(rectlist[1]) + height,
)
sizelist = [ int(x) for x in to_list(v['sourceSize'])]
rect_on_big = big_image.crop(box)
if v['rotated']:
rect_on_big = rect_on_big.rotate(90)
result_image = Image.new('RGBA', sizelist, (0,0,0,0))
if v['rotated']:
result_box=(
( sizelist[0] - height )/2,
( sizelist[1] - width )/2,
( sizelist[0] + height )/2,
( sizelist[1] + width )/2
)
else:
result_box=(
( sizelist[0] - width )/2,
( sizelist[1] - height )/2,
( sizelist[0] + width )/2,
( sizelist[1] + height )/2
)
result_image.paste(rect_on_big, result_box, mask=0)
os.mkdir(file_path)
outfile = (file_path+'/' + k).replace('gift_', '')
print outfile, "generated"
result_image.save(outfile)
```
注:python依赖PIL,本文使用Python Imaging Library 1.1.7 for Python 2.7
官网http://www.pythonware.com/products/pil/安装完后执行
cd C:\Python27\scripts\
pip install pillow
完整示例: https://github.com/l2xin/UnpackSpriteSheet
领取专属 10元无门槛券
私享最新 技术干货