def cal_fid():
"""
与真实图片对比，每张图片通过Illustation2vec feature extractor得到一个1*4096向量
然后计算FID
:return:fid
"""
illust2vec = i2v.make_i2v_with_chainer("illust2vec_ver200.caffemodel")
# 真实图片的特征向量
root = "resize/true/"
paths = os.listdir(root)
true_list = []
for path in paths:
if "jpg" in path:
true_list.append(Image.open(root + path))
res_true = []
for i in range(opt.batch_size):
result_real = illust2vec.extract_feature([true_list[i]])
res_true.append(result_real)
# print(str(i))
true_vec = np.concatenate(tuple(res_true), axis=0)
mu1 = true_vec.mean(axis=0)
sigma1 = np.cov(true_vec, rowvar=False)
print("true_vec_done")
fid1 = temp_cal("resize/fake/", 256, mu1, sigma1)
print("fid1:" + str(fid1))
fid2 = temp_cal("resize/fake2/40/", 64, mu1, sigma1)
print("fid2:" + str(fid2))
fid3 = temp_cal("resize/fake3/40/", 64, mu1, sigma1)
print("fid2:" + str(fid3))
def main(args):
#引数の読み込み
print(args)
src_dir = args.input_dir
out_dir = args.out_dir
threshold = float(args.threshold)
vectorize = args.vectorize
title = os.path.split(src_dir)[0].split('/')[len(os.path.split(src_dir)[0].split('/')) - 2]
if vectorize == "False":
out_file = out_dir + '/' + title + '_tags' + str(threshold) + '_output.csv'
f = open(out_file, 'w')
f.close()
#read model
print("start reading tag model...")
illust2vec = i2v.make_i2v_with_chainer(
"illust2vec_tag_ver200.caffemodel", "tag_list.json")
print("finished reading model")
if vectorize == "True":
out_file = out_dir + '/' + title + '_vecs' + '_output.csv'
f = open(out_file, 'w')
f.close()
#read model
print("start reading vec model...")
#don't have to use tag model and don't have to include tag_list.json
illust2vec = i2v.make_i2v_with_chainer(
"illust2vec_ver200.caffemodel")
print("finished reading model")
# In the case of caffe, please use i2v.make_i2v_with_caffe instead:
# illust2vec = i2v.make_i2v_with_caffe(
# "illust2vec_tag.prototxt", "illust2vec_tag_ver200.caffemodel",
# "tag_list.json")
files = []
jpgs = []
for x in os.listdir(src_dir):
if os.path.isfile(src_dir + x):
files.append(x)
for y in files:
if(y[-4:] == '.jpg'): #ファイル名の後ろ4文字を取り出してそれが.jpgなら
jpgs.append(y) #リストに追加
for file_name in jpgs:
file_path = src_dir + '/' + file_name
vectorize_f(illust2vec, file_path, out_file, threshold, vectorize)
def main():
# Connect to S3
bucket_name = 'pixiv-image-backet'
client = boto3.client('s3')
s3 = boto3.resource('s3')
sbucket = Session().resource('s3').Bucket(bucket_name)
keys = [
obj.key
for obj in sbucket.objects.filter(Prefix='origin_size_face_images')
]
client.download_file(bucket_name, 'image_tag.csv', './image_tag.csv')
image_tag_df = pd.read_csv('image_tag.csv').drop('Unnamed: 0', axis=1)
# Read hair color list
with open('tag.json', 'r') as f:
tag_list = json.load(f)
hair_color_list = tag_list['hair_color']
# Except completed images
completed_image_list = [
'origin_size_face_images/' + img_name
for img_name in list(image_tag_df['image name'])
]
keys = list(set(keys) - set(completed_image_list))
# Load illustration2vec
illust2vec = i2v.make_i2v_with_chainer(
'./i2v/illust2vec_tag_ver200.caffemodel', './i2v/tag_list.json')
# Tag extraction
for key in keys:
print(key)
image_name = key.lstrip('origin_size_face_images/')
client.download_file(bucket_name, key, './tmp_image.png')
image = Image.open('./tmp_image.png')
result = illust2vec.estimate_plausible_tags([image], threshold=0.2)
image_tag_df = image_tag_df.append(pd.Series(
[image_name, None], index=image_tag_df.columns),
ignore_index=True)
for tag, _ in result[0]['general']:
if tag in hair_color_list:
image_tag_df.loc[image_tag_df['image name'] == image_name,
'hair color'] = tag
break
# Uploade to S3
image_tag_df.to_csv('image_tag.csv')
sbucket.upload_file('./image_tag.csv', 'image_tag.csv')
def data_import(width, height):
image = np.empty((0, height, width, DIM), dtype=np.uint8)
list = glob.glob(TRAIN_IMAGE_PATH)
for i in list:
im_reading = Image.open(i).resize((width, height))
im_reading = im_reading.convert("RGB")
im_reading = np.array(im_reading)
print(im_reading.shape)
#im_reading = im_reading.transpose(1,0,2)
print(i)
image = np.append(image, [im_reading], axis=0)
try:
with open('tags.pickle', 'rb') as f:
estimated_tags = pickle.load(f)
print("load from pickle")
except:
try:
with open('illust2vec.pickle', 'rb') as f:
illust2vec = pickle.load(f)
print("pickle i2v")
except:
illust2vec = i2v.make_i2v_with_chainer(
"./i2v/illust2vec_tag_ver200.caffemodel",
"./i2v/tag_list.json")
with open('illust2vec.pickle', 'wb') as f:
pickle.dump(illust2vec, f)
print("new i2v")
batch_size = 5
estimated_tags = np.zeros((0, NUMBER_OF_TAG))
for i in range(math.floor(len(image) / batch_size + 1)):
print(str(i) + "/" + str(math.floor(len(image) / batch_size + 1)))
if len(image) < batch_size * (i + 1):
batch = np.array(image[batch_size * i:])
else:
batch = np.array(image[batch_size * i:batch_size * (i + 1)])
print(estimated_tags.shape)
estimated_tags = np.append(estimated_tags,
illust2vec.extract_feature(batch),
axis=0)
with open('tags.pickle', 'wb') as f:
pickle.dump(estimated_tags, f)
return image, estimated_tags
def get_main_tag(img: Image) -> np.array:
"""
:param img: PIL.Image
:return: tag list
"""
illust2vec_tag = i2v.make_i2v_with_chainer(config.tag_model_path,
config.tag_list_json)
result = illust2vec_tag.estimate_plausible_tags(
[img], threshold=config.tag_threshold)[-1]['general']
feat = []
for each in result:
if each[0] in utils.tag_map.keys():
feat.append(each[0])
return feat
def get_model():
illust2vec = []
p = os.path.abspath(os.path.join("../", os.pardir))
with open(p + "/config.json") as json_data:
config = json.load(json_data)
try:
illust2vec_f = open(config['directories']['ill2vec_ser'], "rb")
illust2vec = pickle.load(illust2vec_f)
illust2vec_f.close()
except IOError:
illust2vec = i2v.make_i2v_with_chainer(
config['directories']['ill2vec_model'],
config['directories']['ill2vec_tag_list'])
save_model = open(config['directories']['ill2vec_ser'], "wb")
pickle.dump(illust2vec, save_model)
save_model.close()
return illust2vec
#!/usr/bin/env python2
import time
import i2v
from PIL import Image
HANDLER_NAME = "i2v_offline"
MAXSIZE = (2500, 2500)
print "Loading model..."
start = time.time()
illust2vec = i2v.make_i2v_with_chainer("data/i2v_model.caffemodel", "data/i2v_tag_list.json")
end = time.time()
print "Done. Took " + str(end - start) + "s"
def run(fname):
def fst(p):
return p[0]
print "Doing inference (i2v - offline)..."
start = time.time()
img = Image.open(fname)
if img.size > MAXSIZE:
print "Image too large. Resizing."
img.thumbnail(MAXSIZE, Image.ANTIALIAS)
res = fst(illust2vec.estimate_plausible_tags([img], threshold=0.5))
import sys
sys.path.append("/Users/KOKI/PerfectMakeGirls/python/i2v")
import i2v
TRAIN_IMAGE_PATH = "/Users/KOKI/Documents/TrainData3/*"
GENERATED_IMAGE_PATH = "/Users/KOKI/Documents/Generated/"
BATCH_SIZE = 200
NUM_EPOCH = 10
DIM = 3
NUMBER_OF_TAG = 1539
try:
with open('illust2vec.pickle', 'r') as f:
illust2vec = pickle.load(f)
except:
illust2vec = i2v.make_i2v_with_chainer(
"./i2v/illust2vec_tag_ver200.caffemodel", "./i2v/tag_list.json")
with open('illust2vec.pickle', 'wb') as f:
pickle.dump(illust2vec, f)
def generator_model(width, height):
model = Sequential()
model.add(
Dense(input_dim=100, output_dim=128, bias_initializer='he_uniform'))
model.add(Activation('tanh'))
model.add(Dense(width * height))
model.add(Activation('tanh'))
model.add(Dense(width * height * 4))
model.add(Activation('tanh'))
model.add(Dense(width * 4 * height * 4 * 4))
model.add(BatchNormalization())
def get_model():
return i2v.make_i2v_with_chainer(
'/usr/src/data/illustration2vec/illust2vec_tag_ver200.caffemodel',
'/usr/src/data/illustration2vec/tag_list.json'
)
import pickle
import codecs
import i2v
import os
from PIL import Image
def getFileList(base):
searchPath = "./" + base + "/*/*.jpg"
flist = glob.glob(searchPath)
flist.sort()
return flist
print("load i2v")
clf = i2v.make_i2v_with_chainer("illust2vec_tag_ver200.caffemodel",
"tag_list.json")
print("load finish")
def I2vTagging(fname, output):
img = Image.open(fname)
data = clf.estimate_plausible_tags([img], threshold=0.0)
fout = codecs.open(output + "pkl", "wb")
pickle.dump(data, fout)
fout.close()
fout = codecs.open(output, "w", encoding="utf-8")
fout.write(str(data))
fout.close()
# print(data)
def __init__(self, model, list):
self.evaluator = i2v.make_i2v_with_chainer(model, list)
def data_import(width, height):
try:
with open('images.pickle', 'rb') as f:
image = pickle.load(f)
print("image load from pickle")
except:
image = np.empty((0, height, width, DIM), dtype=np.uint8)
list = glob.glob(TRAIN_IMAGE_PATH)
for i in list:
im_reading = Image.open(i).resize((width, height))
if im_reading.mode == "RGB":
im_reading = np.array(im_reading)
else:
im_reading = im_reading.convert("RGB")
im_reading = np.array(im_reading)
for j in range(len(im_reading)):
for k in range(len(im_reading[0])):
if np.all(im_reading[j][k] == [71, 112, 76]) or np.all(
im_reading[j][k] == [0, 0, 0]) or np.all(
im_reading[j][k] == [76, 105, 113]):
im_reading[j][k] = [255, 255, 255]
#RGB (71,112,76),(75,105,113)のための例外処理。なんかいい方法があったら言ってくれ
print(i)
#im_reading = im_reading.transpose(1,0,2)
image = np.append(image, [im_reading], axis=0)
with open('images.pickle', 'wb') as f:
pickle.dump(image, f)
print("new image")
try:
with open('tags.pickle', 'rb') as f:
estimated_tags = pickle.load(f)
print("load from pickle")
except:
try:
with open('illust2vec.pickle', 'rb') as f:
illust2vec = pickle.load(f)
print("pickle i2v")
except:
illust2vec = i2v.make_i2v_with_chainer(
"./i2v/illust2vec_tag_ver200.caffemodel",
"./i2v/tag_list.json")
with open('illust2vec.pickle', 'wb') as f:
pickle.dump(illust2vec, f)
print("new i2v")
batch_size = 5
estimated_tags = np.zeros((0, NUMBER_OF_TAG))
for i in range(math.floor(len(image) / batch_size + 1)):
print(str(i) + "/" + str(math.floor(len(image) / batch_size + 1)))
if len(image) < batch_size * (i + 1):
batch = np.array(image[batch_size * i:])
else:
batch = np.array(image[batch_size * i:batch_size * (i + 1)])
print(estimated_tags.shape)
estimated_tags = np.append(estimated_tags,
illust2vec.extract_feature(batch),
axis=0)
with open('tags.pickle', 'wb') as f:
pickle.dump(estimated_tags, f)
return image, estimated_tags
except:
pass
if __name__ == '__main__':
try:
# Tries to load the illustration2vec model
illust2vec_f = open(
"Illustration2Vector/illustration2vec_master/illust2vec.pickle",
"rb")
illust2vec = pickle.load(illust2vec_f)
illust2vec_f.close()
except IOError:
illust2vec = i2v.make_i2v_with_chainer(
"Illustration2Vector/illustration2vec-master/illust2vec_tag_ver200.caffemodel",
"Illustration2Vector/illustration2vec-master/tag_list.json")
save_model = open(
"Illustration2Vector/illustration2vec_master/illust2vec.pickle",
"wb")
pickle.dump(illust2vec, save_model)
save_model.close()
icons = []
screenshots = []
description = ""
# Gets all the content from database
db = sqlite3.connect('./API to download database/app_info_big.db')
c = db.cursor()
explicit_content = []
def __init__(self, model, list):
self.evaluator = i2v.make_i2v_with_chainer(model, list)
import i2v
import cv2
import glob
import os
from imageio import imread
from tqdm import tqdm
import matplotlib.pyplot as plt
import numpy as np
import pickle
# 读取图片路径
images = glob.glob('characters/*.jpg')
print(len(images))
# 加载两个模型
illust2vec = i2v.make_i2v_with_chainer('illust2vec_tag_ver200.caffemodel', 'tag_list.json')
cascade = cv2.CascadeClassifier('lbpcascade_animeface.xml')
OUTPUT_DIR = 'faces/'
if not os.path.exists(OUTPUT_DIR):
os.mkdir(OUTPUT_DIR)
# 提取全部头像，共检测到27772张
num = 0
for x in tqdm(range(len(images))):
img_path = images[x]
image = cv2.imread(img_path)
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
gray = cv2.equalizeHist(gray)
faces = cascade.detectMultiScale(gray, scaleFactor=1.1, minNeighbors=5, minSize=(64, 64))
for (x, y, w, h) in faces:
cx = x + w // 2
#!/usr/bin/env python2
import time
import i2v
from PIL import Image
HANDLER_NAME = "i2v_offline"
MAXSIZE = (2500, 2500)
print "Loading model..."
start = time.time()
illust2vec = i2v.make_i2v_with_chainer("data/i2v_model.caffemodel",
"data/i2v_tag_list.json")
end = time.time()
print "Done. Took " + str(end - start) + "s"
def run(fname):
def fst(p):
return p[0]
print "Doing inference (i2v - offline)..."
start = time.time()
img = Image.open(fname)
if img.size > MAXSIZE:
print "Image too large. Resizing."
img.thumbnail(MAXSIZE, Image.ANTIALIAS)
