专业素材网站的搜图功能：
很多背景墙、墙纸、壁纸、电视墙、装修设计素材网都必配以图搜图。这个以图搜图识图的好处不言而喻，是很多素材网、图片网、三维网等等必备功能。
推荐一款专业的以图搜图系统imgso，它是一个以图搜图专业系统，让你的网站拥有站内设计素材搜图识图功能。
以图搜图在现在应用的非常普遍，是一个非常专业和实用的工具。相对于关键字搜索，以图搜图的方式更加的方便，特别对于特征难以用文字描述的，这个时候图像搜索就能展示出它的强大了。
这款imgso专业搜图采用以Ai智能搜图，神经网络学习底层技术，更有其他丰富的功能设置：
1.拖拽本地图片识图
2.粘贴网络图片地址识图
3.截图粘贴图片识图
4.本地上传图片识图
5.裁图识图
这些都是搜图网站的必备功能，另外Imgso系统插件有增强的功能：
限制登录后搜索：开启该功能后，用户需要登录才能搜索。
搜图分类：开启该功能后，搜索结果只展示你该分类下的素材。
......等等，更多功能设置需要你亲自体验。
需要imgso以图搜图系统，可联系下面演示网的客服。
imgso是wordpress专业插件系统，上传后就能使用。如需搭配其他系统，请和客服联系。
imgso搜图结果部分展示：
底层部分代码：
#以图搜图图片数据分析计算逻辑
from multiprocessing import Pool
from skimage.transform import resize
# Apply transformations to multiple images
def apply_transformer(imgs, transformer, parallel=True):
if parallel:
pool = Pool()
imgs_transform = pool.map(transformer, [img for img in imgs])
pool.close()
pool.join()
else:
imgs_transform = [transformer(img) for img in imgs]
return imgs_transform
# Normalize image data [0, 255] -> [0.0, 1.0]
def normalize_img(img):
return img / 255.
# Resize image
def resize_img(img, shape_resized):
img_resized = resize(img, shape_resized,
anti_aliasing=True,
preserve_range=True)
assert img_resized.shape == shape_resized
return img_resized
# Flatten image
def flatten_img(img):
return img.flatten("C")
"""
import numpy as np
import tensorflow as tf
from src.utils import split
class AutoEncoder():
def __init__(self, modelName, info):
self.modelName = modelName
self.info = info
self.autoencoder = None
self.encoder = None
self.decoder = None
# Train
def fit(self, X, n_epochs=50, batch_size=256):
indices_fracs = split(fracs=[0.9, 0.1], N=len(X), seed=0)
X_train, X_valid = X[indices_fracs[0]], X[indices_fracs[1]]
self.autoencoder.fit(X_train, X_train,
epochs = n_epochs,
batch_size = batch_size,
shuffle = True,
validation_data = (X_valid, X_valid))
# Inference
def predict(self, X):
return self.encoder.predict(X)
# Set neural network architecture
def set_arch(self):
shape_img = self.info["shape_img"]
shape_img_flattened = (np.prod(list(shape_img)),)
# Set encoder and decoder graphs
if self.modelName == "simpleAE":
encode_dim = 128
input = tf.keras.Input(shape=shape_img_flattened)
encoded = tf.keras.layers.Dense(encode_dim, activation='relu')(input)
decoded = tf.keras.layers.Dense(shape_img_flattened[0], activation='sigmoid')(encoded)
elif self.modelName == "convAE":
n_hidden_1, n_hidden_2, n_hidden_3 = 16, 8, 8
convkernel = (3, 3) # convolution kernel
poolkernel = (2, 2) # pooling kernel
input = tf.keras.layers.Input(shape=shape_img)
x = tf.keras.layers.Conv2D(n_hidden_1, convkernel, activation='relu', padding='same')(input)
x = tf.keras.layers.MaxPooling2D(poolkernel, padding='same')(x)
x = tf.keras.layers.Conv2D(n_hidden_2, convkernel, activation='relu', padding='same')(x)
x = tf.keras.layers.MaxPooling2D(poolkernel, padding='same')(x)
x = tf.keras.layers.Conv2D(n_hidden_3, convkernel, activation='relu', padding='same')(x)
encoded = tf.keras.layers.MaxPooling2D(poolkernel, padding='same')(x)
x = tf.keras.layers.Conv2D(n_hidden_3, convkernel, activation='relu', padding='same')(encoded)
x = tf.keras.layers.UpSampling2D(poolkernel)(x)
x = tf.keras.layers.Conv2D(n_hidden_2, convkernel, activation='relu', padding='same')(x)
x = tf.keras.layers.UpSampling2D(poolkernel)(x)
x = tf.keras.layers.Conv2D(n_hidden_1, convkernel, activation='relu')(x)
x = tf.keras.layers.UpSampling2D(poolkernel)(x)
decoded = tf.keras.layers.Conv2D(shape_img[2], convkernel, activation='sigmoid', padding='same')(x)
else:
raise Exception("Invalid model name given!")
# Create autoencoder model
autoencoder = tf.keras.Model(input, decoded)
input_autoencoder_shape = autoencoder.layers[0].input_shape[1:]
output_autoencoder_shape = autoencoder.layers[-1].output_shape[1:]
# Create encoder model
encoder = tf.keras.Model(input, encoded) # set encoder
input_encoder_shape = encoder.layers[0].input_shape[1:]
output_encoder_shape = encoder.layers[-1].output_shape[1:]
# Create decoder model
decoded_input = tf.keras.Input(shape=output_encoder_shape)
if self.modelName == 'simpleAE':
decoded_output = autoencoder.layers[-1](decoded_input) # single layer
elif self.modelName == 'convAE':
decoded_output = autoencoder.layers[-7](decoded_input) # Conv2D
decoded_output = autoencoder.layers[-6](decoded_output) # UpSampling2D
decoded_output = autoencoder.layers[-5](decoded_output) # Conv2D
decoded_output = autoencoder.layers[-4](decoded_output) # UpSampling2D
decoded_output = autoencoder.layers[-3](decoded_output) # Conv2D
decoded_output = autoencoder.layers[-2](decoded_output) # UpSampling2D
decoded_output = autoencoder.layers[-1](decoded_output) # Conv2D
else:
raise Exception("Invalid model name given!")
decoder = tf.keras.Model(decoded_input, decoded_output)
decoder_input_shape = decoder.layers[0].input_shape[1:]
decoder_output_shape = decoder.layers[-1].output_shape[1:]
# Generate summaries
print("\nautoencoder.summary():")
print(autoencoder.summary())
print("\nencoder.summary():")
print(encoder.summary())
print("\ndecoder.summary():")
print(decoder.summary())
# Assign models
self.autoencoder = autoencoder
self.encoder = encoder
self.decoder = decoder
# Compile
def compile(self, loss="binary_crossentropy", optimizer="adam"):
self.autoencoder.compile(optimizer=optimizer, loss=loss)
# Load model architecture and weights
def load_models(self, loss="binary_crossentropy", optimizer="adam"):
print("Loading models...")
self.autoencoder = tf.keras.models.load_model(self.info["autoencoderFile"])
self.encoder = tf.keras.models.load_model(self.info["encoderFile"])
self.decoder = tf.keras.models.load_model(self.info["decoderFile"])
self.autoencoder.compile(optimizer=optimizer, loss=loss)
self.encoder.compile(optimizer=optimizer, loss=loss)
self.decoder.compile(optimizer=optimizer, loss=loss)
# Save model architecture and weights to file
def save_models(self):
print("Saving models...")
self.autoencoder.save(self.info["autoencoderFile"])
self.encoder.save(self.info["encoderFile"])
self.decoder.save(self.info["decoderFile"])
扫一扫
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
