opensourcegames/code/html/osgameclones_download_images_create_collage.py

202 lines
6.9 KiB
Python

"""
Downloads images from games, stored in the osgameclones-database, then creates a collage of them.
"""
import ruamel.yaml as yaml
import os
import requests
from PIL import Image
from io import BytesIO
import numpy as np
from progress.bar import IncrementalBar
def download_images():
# import the osgameclones data
path = os.path.realpath(os.path.join(root_path, os.path.pardir, 'osgameclones.git', 'games'))
files = os.listdir(path)
# iterate over all yaml files in osgameclones/data folder and load contents
entries = []
for file in files:
# read yaml
with open(os.path.join(path, file), 'r', encoding='utf-8') as stream:
try:
_ = yaml.safe_load(stream)
except Exception as exc:
print(file)
raise exc
# add to entries
entries.extend(_)
print('imported {} entries'.format(len(entries)))
# collect all image informations
images = []
for entry in entries:
if 'images' in entry:
images.extend(entry['images'])
print('contain {} image links'.format(len(images)))
# download them all
for url in images:
name = "".join(x for x in url[5:] if (x.isalnum() or x in '._-'))
outfile = os.path.join(download_path, name)
if not os.path.isfile(outfile):
try:
r = requests.get(url, headers={'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; WOW64)'},
timeout=20, allow_redirects=True)
if r.status_code == requests.codes.ok:
im = Image.open(BytesIO(r.content))
im.save(outfile)
print('saved {}'.format(url))
except:
pass
def downsize_images():
scale_factor = 10
for file in os.listdir(download_path):
file_path = os.path.join(download_path, file)
if not os.path.isfile(file_path):
continue
outfile = os.path.join(downsized_path, file[:-4]+'.png') # losless storage of downsize image
if os.path.isfile(outfile):
continue
im = Image.open(file_path)
if im.mode != 'RGB':
print('{} - {}'.format(file, im.mode))
continue
width = im.width
height = im.height
if width < target_width * scale_factor or height < target_height * scale_factor:
continue
box = [(width-target_width*scale_factor)/2, (height-target_height*scale_factor)/2, target_width * scale_factor, target_height * scale_factor]
box[2] += box[0]
box[3] += box[1]
im_resized = im.resize((target_width, target_height), resample=Image.LANCZOS, box=box)
im_resized.save(outfile)
print('saved {}'.format(file))
def assemble_collage():
print('start assembling collage')
# load all from downsized path
files = os.listdir(downsized_path)
files = [file for file in files if os.path.isfile(os.path.join(downsized_path, file))]
images = []
bar = IncrementalBar('Loading', max=len(files))
for file in files:
im = Image.open(os.path.join(downsized_path, file))
im = np.asarray(im)
images.append(im)
bar.next()
bar.finish()
# compute total amount of light in each image and only keep the N brightest
images = [(np.sum(image), image) for image in images]
images.sort(key=lambda x: x[0], reverse=True)
images = images[:N]
images = [x[1] for x in images]
# compute the average color in each quadrant
Cx = int(target_height / 2)
Cy = int(target_width / 2)
U = [np.mean(image[:Cx, :, :], axis=(1, 2)) for image in images]
D = [np.mean(image[Cx:, :, :], axis=(1, 2)) for image in images]
R = [np.mean(image[:, :Cy, :], axis=(1, 2)) for image in images]
L = [np.mean(image[:, Cy:, :], axis=(1, 2)) for image in images]
# initially just sort them in randomly
map = np.random.permutation(N).reshape((Nx, Ny))
# optimize neighbors with a stochastic metropolis algorithm
Ni = 500000
T = np.linspace(150, 2, Ni)
A = np.zeros((Ni, 1))
u = lambda x: (x + 1) % Nx
d = lambda x: (x - 1) % Nx
r = lambda x: (x + 1) % Ny
l = lambda x: (x - 1) % Ny
score = lambda i1, j1, i2, j2: np.linalg.norm(U[map[i1, j1]] - D[map[u(i2), j2]]) + np.linalg.norm(D[map[i1, j1]] - U[map[d(i2), j2]]) + np.linalg.norm(L[map[i1, j1]] - R[map[i2, l(j2)]]) + np.linalg.norm(R[map[i1, j1]] - L[map[i2, r(j2)]])
bar = IncrementalBar('Optimization', max=Ni)
for ai in range(Ni):
# get two non-equal random locations
i1 = np.random.randint(Nx)
j1 = np.random.randint(Ny)
while True:
i2 = np.random.randint(Nx)
j2 = np.random.randint(Ny)
if i1 != i2 or j1 != j2:
break
# compute score
x = score(i1, j1, i1, j1) - score(i1, j1, i2, j2) + score(i2, j2, i2, j2) - score(i2, j2, i1, j1)
# exchange
# if x < 0:
# if x > 0:
if x > 0 or np.exp(x / T[ai]) > np.random.uniform():
map[i1, j1], map[i2, j2] = map[i2, j2], map[i1, j1]
A[ai] = 1
bar.next()
bar.finish()
# time evolution of acceptance rate
Nc = int(np.floor(Ni / 20))
for ai in range(20):
print('{}: {}'.format(ai, np.mean(A[ai*Nc:(ai+1)*Nc])))
# shift brightest to center
B = np.zeros((Nx, Ny))
for i in range(Nx):
for j in range(Ny):
B[i, j] = np.sum(images[map[i, j]])
sk = np.array([0.25, 0.5, 1, 0.5, 0.25])
# convolve in 1D along all rows and all columns
for i in range(Nx):
B[i, :] = np.convolve(B[i, :], sk, mode='same')
for j in range(Ny):
B[:, j] = np.convolve(B[:, j], sk, mode='same')
cx, cy = np.unravel_index(np.argmax(B), B.shape)
map = np.roll(map, (int(Nx/2-cx), int(Ny/2-cy)), axis=(0, 1))
# assemble image
final = np.zeros((Nx * target_height, Ny * target_width, 3), dtype=np.uint8)
for i in range(Nx):
for j in range(Ny):
final[i*target_height:(i+1)*target_height, j*target_width:(j+1)*target_width] = images[map[i, j]]
# convert back to pillow image and save
im = Image.fromarray(final)
im.save(output_file)
if __name__ == "__main__":
target_height = 60
target_width = 80
Nx = 12 # vertical
Ny = 18 # horizontal
N = Nx * Ny
# paths
root_path = os.path.realpath(os.path.join(os.path.dirname(__file__), os.path.pardir))
download_path = os.path.join(root_path, 'code', '', 'images-download')
downsized_path = os.path.join(download_path, 'downsized')
output_file = os.path.join(root_path, 'code', '', 'collage_games.jpg')
if not os.path.exists(download_path):
os.mkdir(download_path)
if not os.path.exists(downsized_path):
os.mkdir(downsized_path)
# download files
# download_images()
# downsize downloaded images
# downsize_images()
# assemble collage
assemble_collage()