自然语言图像搜索与双编码器

pip install -q -U tensorflow-hub tensorflow-text tensorflow-addons

import os
import collections
import json
import numpy as np
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers
import tensorflow_hub as hub
import tensorflow_text as text
import tensorflow_addons as tfa
import matplotlib.pyplot as plt
import matplotlib.image as mpimg
from tqdm import tqdm

# 抑制tf.hub警告
tf.get_logger().setLevel("ERROR")

root_dir = "datasets"
annotations_dir = os.path.join(root_dir, "annotations")
images_dir = os.path.join(root_dir, "train2014")
tfrecords_dir = os.path.join(root_dir, "tfrecords")
annotation_file = os.path.join(annotations_dir, "captions_train2014.json")

# 下载标题注释文件
if not os.path.exists(annotations_dir):
    annotation_zip = tf.keras.utils.get_file(
        "captions.zip",
        cache_dir=os.path.abspath("."),
        origin="http://images.cocodataset.org/annotations/annotations_trainval2014.zip",
        extract=True,
    )
    os.remove(annotation_zip)

# 下载图像文件
if not os.path.exists(images_dir):
    image_zip = tf.keras.utils.get_file(
        "train2014.zip",
        cache_dir=os.path.abspath("."),
        origin="http://images.cocodataset.org/zips/train2014.zip",
        extract=True,
    )
    os.remove(image_zip)

print("数据集已成功下载并提取。")

with open(annotation_file, "r") as f:
    annotations = json.load(f)["annotations"]

image_path_to_caption = collections.defaultdict(list)
for element in annotations:
    caption = f"{element['caption'].lower().rstrip('.')}"
    image_path = images_dir + "/COCO_train2014_" + "%012d.jpg" % (element["image_id"])
    image_path_to_caption[image_path].append(caption)

image_paths = list(image_path_to_caption.keys())
print(f"图像数量: {len(image_paths)}")

train_size = 30000
valid_size = 5000
captions_per_image = 2
images_per_file = 2000

train_image_paths = image_paths[:train_size]
num_train_files = int(np.ceil(train_size / images_per_file))
train_files_prefix = os.path.join(tfrecords_dir, "train")

valid_image_paths = image_paths[-valid_size:]
num_valid_files = int(np.ceil(valid_size / images_per_file))
valid_files_prefix = os.path.join(tfrecords_dir, "valid")

tf.io.gfile.makedirs(tfrecords_dir)


def bytes_feature(value):
    return tf.train.Feature(bytes_list=tf.train.BytesList(value=[value]))


def create_example(image_path, caption):
    feature = {
        "caption": bytes_feature(caption.encode()),
        "raw_image": bytes_feature(tf.io.read_file(image_path).numpy()),
    }
    return tf.train.Example(features=tf.train.Features(feature=feature))


def write_tfrecords(file_name, image_paths):
    caption_list = []
    image_path_list = []
    for image_path in image_paths:
        captions = image_path_to_caption[image_path][:captions_per_image]
        caption_list.extend(captions)
        image_path_list.extend([image_path] * len(captions))

    with tf.io.TFRecordWriter(file_name) as writer:
        for example_idx in range(len(image_path_list)):
            example = create_example(
                image_path_list[example_idx], caption_list[example_idx]
            )
            writer.write(example.SerializeToString())
    return example_idx + 1


def write_data(image_paths, num_files, files_prefix):
    example_counter = 0
    for file_idx in tqdm(range(num_files)):
        file_name = files_prefix + "-%02d.tfrecord" % (file_idx)
        start_idx = images_per_file * file_idx
        end_idx = start_idx + images_per_file
        example_counter += write_tfrecords(file_name, image_paths[start_idx:end_idx])
    return example_counter


train_example_count = write_data(train_image_paths, num_train_files, train_files_prefix)
print(f"{train_example_count} training examples were written to tfrecord files.")

valid_example_count = write_data(valid_image_paths, num_valid_files, valid_files_prefix)
print(f"{valid_example_count} evaluation examples were written to tfrecord files.")

feature_description = {
    "caption": tf.io.FixedLenFeature([], tf.string),
    "raw_image": tf.io.FixedLenFeature([], tf.string),
}


def read_example(example):
    features = tf.io.parse_single_example(example, feature_description)
    raw_image = features.pop("raw_image")
    features["image"] = tf.image.resize(
        tf.image.decode_jpeg(raw_image, channels=3), size=(299, 299)
    )
    return features


def get_dataset(file_pattern, batch_size):

    return (
        tf.data.TFRecordDataset(tf.data.Dataset.list_files(file_pattern))
        .map(
            read_example,
            num_parallel_calls=tf.data.AUTOTUNE,
            deterministic=False,
        )
        .shuffle(batch_size * 10)
        .prefetch(buffer_size=tf.data.AUTOTUNE)
        .batch(batch_size)
    )

def project_embeddings(
    embeddings, num_projection_layers, projection_dims, dropout_rate
):
    projected_embeddings = layers.Dense(units=projection_dims)(embeddings)
    for _ in range(num_projection_layers):
        x = tf.nn.gelu(projected_embeddings)
        x = layers.Dense(projection_dims)(x)
        x = layers.Dropout(dropout_rate)(x)
        x = layers.Add()([projected_embeddings, x])
        projected_embeddings = layers.LayerNormalization()(x)
    return projected_embeddings

def create_vision_encoder(
    num_projection_layers, projection_dims, dropout_rate, trainable=False
):
    # 加载预训练的 Xception 模型，用作基础编码器。
    xception = keras.applications.Xception(
        include_top=False, weights="imagenet", pooling="avg"
    )
    # 设置基础编码器的可训练性。
    for layer in xception.layers:
        layer.trainable = trainable
    # 接收图像作为输入。
    inputs = layers.Input(shape=(299, 299, 3), name="image_input")
    # 预处理输入图像。
    xception_input = tf.keras.applications.xception.preprocess_input(inputs)
    # 使用 xception 模型生成图像的嵌入。
    embeddings = xception(xception_input)
    # 对模型生成的嵌入进行投影。
    outputs = project_embeddings(
        embeddings, num_projection_layers, projection_dims, dropout_rate
    )
    # 创建视觉编码器模型。
    return keras.Model(inputs, outputs, name="vision_encoder")

def create_text_encoder(
    num_projection_layers, projection_dims, dropout_rate, trainable=False
):
    # 加载 BERT 预处理模块。
    preprocess = hub.KerasLayer(
        "https://tfhub.dev/tensorflow/bert_en_uncased_preprocess/2",
        name="text_preprocessing",
    )
    # 加载用于作为基础编码器的预训练 BERT 模型。
    bert = hub.KerasLayer(
        "https://tfhub.dev/tensorflow/small_bert/bert_en_uncased_L-4_H-512_A-8/1",
        "bert",
    )
    # 设置基础编码器的可训练性。
    bert.trainable = trainable
    # 接收文本作为输入。
    inputs = layers.Input(shape=(), dtype=tf.string, name="text_input")
    # 对文本进行预处理。
    bert_inputs = preprocess(inputs)
    # 使用 BERT 模型为预处理文本生成嵌入。
    embeddings = bert(bert_inputs)["pooled_output"]
    # 对模型生成的嵌入进行投影。
    outputs = project_embeddings(
        embeddings, num_projection_layers, projection_dims, dropout_rate
    )
    # 创建文本编码器模型。
    return keras.Model(inputs, outputs, name="text_encoder")

class DualEncoder(keras.Model):
    def __init__(self, text_encoder, image_encoder, temperature=1.0, **kwargs):
        super().__init__(**kwargs)
        self.text_encoder = text_encoder
        self.image_encoder = image_encoder
        self.temperature = temperature
        self.loss_tracker = keras.metrics.Mean(name="loss")

    @property
    def metrics(self):
        return [self.loss_tracker]

    def call(self, features, training=False):
        # 将每个编码器放在单独的 GPU 上（如果可用）。
        # 如果 GPU 少于 2 个，TF 将回退到可用设备。
        with tf.device("/gpu:0"):
            # 获取标题的嵌入。
            caption_embeddings = text_encoder(features["caption"], training=training)
        with tf.device("/gpu:1"):
            # 获取图像的嵌入。
            image_embeddings = vision_encoder(features["image"], training=training)
        return caption_embeddings, image_embeddings

    def compute_loss(self, caption_embeddings, image_embeddings):
        # logits[i][j] 是 caption_i 和 image_j 的点相似度。
        logits = (
            tf.matmul(caption_embeddings, image_embeddings, transpose_b=True)
            / self.temperature
        )
        # images_similarity[i][j] 是 image_i 和 image_j 的点相似度。
        images_similarity = tf.matmul(
            image_embeddings, image_embeddings, transpose_b=True
        )
        # captions_similarity[i][j] 是 caption_i 和 caption_j 的点相似度。
        captions_similarity = tf.matmul(
            caption_embeddings, caption_embeddings, transpose_b=True
        )
        # targets[i][j] = caption_i 和 caption_j 的平均点相似度以及 image_i 和 image_j 的点相似度。
        targets = keras.activations.softmax(
            (captions_similarity + images_similarity) / (2 * self.temperature)
        )
        # 使用交叉熵计算标题的损失
        captions_loss = keras.losses.categorical_crossentropy(
            y_true=targets, y_pred=logits, from_logits=True
        )
        # 使用交叉熵计算图像的损失
        images_loss = keras.losses.categorical_crossentropy(
            y_true=tf.transpose(targets), y_pred=tf.transpose(logits), from_logits=True
        )
        # 返回批处理的损失平均值。
        return (captions_loss + images_loss) / 2

    def train_step(self, features):
        with tf.GradientTape() as tape:
            # 前向传递
            caption_embeddings, image_embeddings = self(features, training=True)
            loss = self.compute_loss(caption_embeddings, image_embeddings)
        # 后向传递
        gradients = tape.gradient(loss, self.trainable_variables)
        self.optimizer.apply_gradients(zip(gradients, self.trainable_variables))
        # 监测损失
        self.loss_tracker.update_state(loss)
        return {"loss": self.loss_tracker.result()}

    def test_step(self, features):
        caption_embeddings, image_embeddings = self(features, training=False)
        loss = self.compute_loss(caption_embeddings, image_embeddings)
        self.loss_tracker.update_state(loss)
        return {"loss": self.loss_tracker.result()}

num_epochs = 5  # 实际上，训练至少需要30个epoch
batch_size = 256

vision_encoder = create_vision_encoder(
    num_projection_layers=1, projection_dims=256, dropout_rate=0.1
)
text_encoder = create_text_encoder(
    num_projection_layers=1, projection_dims=256, dropout_rate=0.1
)
dual_encoder = DualEncoder(text_encoder, vision_encoder, temperature=0.05)
dual_encoder.compile(
    optimizer=tfa.optimizers.AdamW(learning_rate=0.001, weight_decay=0.001)
)

print(f"Number of GPUs: {len(tf.config.list_physical_devices('GPU'))}")
print(f"Number of examples (caption-image pairs): {train_example_count}")
print(f"Batch size: {batch_size}")
print(f"Steps per epoch: {int(np.ceil(train_example_count / batch_size))}")
train_dataset = get_dataset(os.path.join(tfrecords_dir, "train-*.tfrecord"), batch_size)
valid_dataset = get_dataset(os.path.join(tfrecords_dir, "valid-*.tfrecord"), batch_size)
# 创建学习率调度器回调。
reduce_lr = keras.callbacks.ReduceLROnPlateau(
    monitor="val_loss", factor=0.2, patience=3
)
# 创建提前停止回调。
early_stopping = tf.keras.callbacks.EarlyStopping(
    monitor="val_loss", patience=5, restore_best_weights=True
)
history = dual_encoder.fit(
    train_dataset,
    epochs=num_epochs,
    validation_data=valid_dataset,
    callbacks=[reduce_lr, early_stopping],
)
print("训练完成。正在保存视觉和文本编码器...")
vision_encoder.save("vision_encoder")
text_encoder.save("text_encoder")
print("模型已保存。")

plt.plot(history.history["loss"])
plt.plot(history.history["val_loss"])
plt.ylabel("损失")
plt.xlabel("周期")
plt.legend(["训练", "验证"], loc="upper right")
plt.show()

print("加载视觉和文本编码器...")
vision_encoder = keras.models.load_model("vision_encoder")
text_encoder = keras.models.load_model("text_encoder")
print("模型已加载。")


def read_image(image_path):
    image_array = tf.image.decode_jpeg(tf.io.read_file(image_path), channels=3)
    return tf.image.resize(image_array, (299, 299))


print(f"为{len(image_paths)}张图像生成嵌入...")
image_embeddings = vision_encoder.predict(
    tf.data.Dataset.from_tensor_slices(image_paths).map(read_image).batch(batch_size),
    verbose=1,
)
print(f"图像嵌入形状: {image_embeddings.shape}.")

def find_matches(image_embeddings, queries, k=9, normalize=True):
    # 获取查询的嵌入。
    query_embedding = text_encoder(tf.convert_to_tensor(queries))
    # 标准化查询和图像嵌入。
    if normalize:
        image_embeddings = tf.math.l2_normalize(image_embeddings, axis=1)
        query_embedding = tf.math.l2_normalize(query_embedding, axis=1)
    # 计算查询与图像嵌入之间的点积。
    dot_similarity = tf.matmul(query_embedding, image_embeddings, transpose_b=True)
    # 检索前 k 个索引。
    results = tf.math.top_k(dot_similarity, k).indices.numpy()
    # 返回匹配的图像路径。
    return [[image_paths[idx] for idx in indices] for indices in results]

query = "一家人在沙滩上站在海洋旁边，旁边有一个冲浪板"
matches = find_matches(image_embeddings, [query], normalize=True)[0]

plt.figure(figsize=(20, 20))
for i in range(9):
    ax = plt.subplot(3, 3, i + 1)
    plt.imshow(mpimg.imread(matches[i]))
    plt.axis("off")

def compute_top_k_accuracy(image_paths, k=100):
    hits = 0
    num_batches = int(np.ceil(len(image_paths) / batch_size))
    for idx in tqdm(range(num_batches)):
        start_idx = idx * batch_size
        end_idx = start_idx + batch_size
        current_image_paths = image_paths[start_idx:end_idx]
        queries = [
            image_path_to_caption[image_path][0] for image_path in current_image_paths
        ]
        result = find_matches(image_embeddings, queries, k)
        hits += sum(
            [
                image_path in matches
                for (image_path, matches) in list(zip(current_image_paths, result))
            ]
        )

    return hits / len(image_paths)


print("评分训练数据...")
train_accuracy = compute_top_k_accuracy(train_image_paths)
print(f"训练准确率: {round(train_accuracy * 100, 3)}%")

print("评分评估数据...")
eval_accuracy = compute_top_k_accuracy(image_paths[train_size:])
print(f"评估准确率: {round(eval_accuracy * 100, 3)}%")