有效地讀取圖像，對比opencv、PIL、turbojpeg、lmdb、tfrecords

opencv和PIL都是很常見的圖像處理庫了，就不介紹了，主要介紹后面三個：

turbojpeg：libjpeg-turbo的python包裝器，用於jpeg圖像的解碼和編碼。

基本用法：

import cv2
from turbojpeg import TurboJPEG, TJPF_GRAY, TJSAMP_GRAY, TJFLAG_PROGRESSIVE
# using default library installation
jpeg = TurboJPEG()

# decoding input.jpg to BGR array
in_file = open('input.jpg', 'rb')
bgr_array = jpeg.decode(in_file.read())
in_file.close()
cv2.imshow('bgr_array', bgr_array)
cv2.waitKey(0)

更多信息參考：https://www.cnpython.com/pypi/pyturbojpeg

lmdb：LMDB的全稱是Lightning Memory-Mapped Database(快如閃電的內存映射數據庫)。LMDB文件可以同時由多個進程打開，具有極高的數據存取速度，訪問簡單，不需要運行單獨的數據庫管理進程，只要在訪問數據的代碼里引用LMDB庫，訪問時給文件路徑即可。讓系統訪問大量小文件的開銷很大，而LMDB使用內存映射的方式訪問文件，使得文件內尋址的開銷非常小，使用指針運算就能實現。數據庫單文件還能減少數據集復制/傳輸過程的開銷。

基本用法：

# -*- coding: utf-8 -*-
import lmdb
  
# 如果train文件夾下沒有data.mbd或lock.mdb文件，則會生成一個空的，如果有，不會覆蓋
# map_size定義最大儲存容量，單位是kb，以下定義1TB容量
env = lmdb.open("./train"，map_size=1099511627776)
env.close()

更多信息參考：https://blog.csdn.net/weixin_41874599/article/details/86631186

tfrecords：frecords是一種二進制編碼的文件格式，tensorflow專用。 能將任意數據轉換為tfrecords。 更好的利用內存，更方便復制和移動，並且不需要單獨的標簽文件。

將圖像轉換為lmdb格式的數據：

import os
from argparse import ArgumentParser

import cv2
import lmdb
import numpy as np

from tools import get_images_paths


def store_many_lmdb(images_list, save_path):

    num_images = len(images_list)  # number of images in our folder

    file_sizes = [os.path.getsize(item) for item in images_list]  # all file sizes
    max_size_index = np.argmax(file_sizes)  # the maximum file size index

    # maximum database size in bytes
    map_size = num_images * cv2.imread(images_list[max_size_index]).nbytes * 10

    env = lmdb.open(save_path, map_size=map_size)  # create lmdb environment

    with env.begin(write=True) as txn:  # start writing to environment
        for i, image in enumerate(images_list):
            with open(image, "rb") as file:
                data = file.read()  # read image as bytes
                key = f"{i:08}"  # get image key
                txn.put(key.encode("ascii"), data)  # put the key-value into database

    env.close()  # close the environment


if __name__ == "__main__":
    parser = ArgumentParser()
    parser.add_argument(
        "--path",
        "-p",
        type=str,
        required=True,
        help="path to the images folder to collect",
    )
    parser.add_argument(
        "--output",
        "-o",
        type=str,
        required=True,
        help='path to the output environment directory file i.e. "path/to/folder/env/"',
    )

    args = parser.parse_args()
    if not os.path.exists(args.output):
        os.makedirs(args.output)

    images = get_images_paths(args.path)
    store_many_lmdb(images, args.output)

將圖像轉換為tfrecords格式的數據：

import os
from argparse import ArgumentParser

import tensorflow as tf

from tools import get_images_paths


def _byte_feature(value):
    """Convert string / byte into bytes_list."""
    if isinstance(value, type(tf.constant(0))):
        value = value.numpy()  # BytesList can't unpack string from EagerTensor.
    return tf.train.Feature(bytes_list=tf.train.BytesList(value=[value]))


def _int64_feature(value):
    """Convert bool / enum / int / uint into int64_list."""
    return tf.train.Feature(int64_list=tf.train.Int64List(value=[value]))


def image_example(image_string, label):
    feature = {
        "label": _int64_feature(label),
        "image_raw": _byte_feature(image_string),
    }
    return tf.train.Example(features=tf.train.Features(feature=feature))


def store_many_tfrecords(images_list, save_file):

    assert save_file.endswith(
        ".tfrecords",
    ), 'File path is wrong, it should contain "*myname*.tfrecords"'

    directory = os.path.dirname(save_file)
    if not os.path.exists(directory):
        os.makedirs(directory)

    with tf.io.TFRecordWriter(save_file) as writer:  # start writer
        for label, filename in enumerate(images_list):  # cycle by each image path
            image_string = open(filename, "rb").read()  # read the image as bytes string
            tf_example = image_example(
                image_string, label,
            )  # save the data as tf.Example object
            writer.write(tf_example.SerializeToString())  # and write it into database


if __name__ == "__main__":
    parser = ArgumentParser()
    parser.add_argument(
        "--path",
        "-p",
        type=str,
        required=True,
        help="path to the images folder to collect",
    )
    parser.add_argument(
        "--output",
        "-o",
        type=str,
        required=True,
        help='path to the output tfrecords file i.e. "path/to/folder/myname.tfrecords"',
    )

    args = parser.parse_args()
    image_paths = get_images_paths(args.path)
    store_many_tfrecords(image_paths, args.output)

使用不同的方式讀取圖像，同時默認是以BGR的格式讀取：

import os
from abc import abstractmethod
from timeit import default_timer as timer

import cv2
import lmdb
import numpy as np
import tensorflow as tf
from PIL import Image
from turbojpeg import TurboJPEG

os.environ["TF_CPP_MIN_LOG_LEVEL"] = "3"


class ImageLoader:
    extensions: tuple = (".png", ".jpg", ".jpeg", ".tiff", ".bmp", ".gif", ".tfrecords")

    def __init__(self, path: str, mode: str = "BGR"):
        self.path = path
        self.mode = mode
        self.dataset = self.parse_input(self.path)
        self.sample_idx = 0

    def parse_input(self, path):

        # single image or tfrecords file
        if os.path.isfile(path):
            assert path.lower().endswith(
                self.extensions,
            ), f"Unsupportable extension, please, use one of {self.extensions}"
            return [path]

        if os.path.isdir(path):
            # lmdb environment
            if any([file.endswith(".mdb") for file in os.listdir(path)]):
                return path
            else:
                # folder with images
                paths = [os.path.join(path, image) for image in os.listdir(path)]
                return paths

    def __iter__(self):
        self.sample_idx = 0
        return self

    def __len__(self):
        return len(self.dataset)

    @abstractmethod
    def __next__(self):
        pass


class CV2Loader(ImageLoader):
    def __next__(self):
        start = timer()
        path = self.dataset[self.sample_idx]  # get image path by index from the dataset
        image = cv2.imread(path)  # read the image
        full_time = timer() - start
        if self.mode == "RGB":
            start = timer()
            image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)  # change color mode
            full_time += timer() - start
        self.sample_idx += 1
        return image, full_time


class PILLoader(ImageLoader):
    def __next__(self):
        start = timer()
        path = self.dataset[self.sample_idx]  # get image path by index from the dataset
        image = np.asarray(Image.open(path))  # read the image as numpy array
        full_time = timer() - start
        if self.mode == "BGR":
            start = timer()
            image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)  # change color mode
            full_time += timer() - start
        self.sample_idx += 1
        return image, full_time


class TurboJpegLoader(ImageLoader):
    def __init__(self, path, **kwargs):
        super(TurboJpegLoader, self).__init__(path, **kwargs)
        self.jpeg_reader = TurboJPEG()  # create TurboJPEG object for image reading

    def __next__(self):
        start = timer()
        file = open(self.dataset[self.sample_idx], "rb")  # open the input file as bytes
        full_time = timer() - start
        if self.mode == "RGB":
            mode = 0
        elif self.mode == "BGR":
            mode = 1
        start = timer()
        image = self.jpeg_reader.decode(file.read(), mode)  # decode raw image
        full_time += timer() - start
        self.sample_idx += 1
        return image, full_time


class LmdbLoader(ImageLoader):
    def __init__(self, path, **kwargs):
        super(LmdbLoader, self).__init__(path, **kwargs)
        self.path = path
        self._dataset_size = 0
        self.dataset = self.open_database()

    # we need to open the database to read images from it
    def open_database(self):
        lmdb_env = lmdb.open(self.path)  # open the environment by path
        lmdb_txn = lmdb_env.begin()  # start reading
        lmdb_cursor = lmdb_txn.cursor()  # create cursor to iterate through the database
        self._dataset_size = lmdb_env.stat()[
            "entries"
        ]  # get number of items in full dataset
        return lmdb_cursor

    def __iter__(self):
        self.dataset.first()  # return the cursor to the first database element
        return self

    def __next__(self):
        start = timer()
        raw_image = self.dataset.value()  # get raw image
        image = np.frombuffer(raw_image, dtype=np.uint8)  # convert it to numpy
        image = cv2.imdecode(image, cv2.IMREAD_COLOR)  # decode image
        full_time = timer() - start
        if self.mode == "RGB":
            start = timer()
            image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
            full_time += timer() - start
        start = timer()
        self.dataset.next()  # step to the next element in database
        full_time += timer() - start
        return image, full_time

    def __len__(self):
        return self._dataset_size  # get dataset length


class TFRecordsLoader(ImageLoader):
    def __init__(self, path, **kwargs):
        super(TFRecordsLoader, self).__init__(path, **kwargs)
        self._dataset = self.open_database()

    def open_database(self):
        def _parse_image_function(example_proto):
            return tf.io.parse_single_example(example_proto, image_feature_description)

        # dataset structure description
        image_feature_description = {
            "label": tf.io.FixedLenFeature([], tf.int64),
            "image_raw": tf.io.FixedLenFeature([], tf.string),
        }
        raw_image_dataset = tf.data.TFRecordDataset(self.path)  # open dataset by path
        parsed_image_dataset = raw_image_dataset.map(
            _parse_image_function,
        )  # parse dataset using structure description

        return parsed_image_dataset

    def __iter__(self):
        self.dataset = self._dataset.as_numpy_iterator()
        return self

    def __next__(self):
        start = timer()
        value = next(self.dataset)[
            "image_raw"
        ]  # step to the next element in database and get new image
        image = tf.image.decode_jpeg(value).numpy()  # decode raw image
        full_time = timer() - start
        if self.mode == "BGR":
            start = timer()
            image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
            full_time += timer() - start
        return image, full_time

    def __len__(self):
        return self._dataset.reduce(
            np.int64(0), lambda x, _: x + 1,
        ).numpy()  # get dataset length


methods = {
    "cv2": CV2Loader,
    "pil": PILLoader,
    "turbojpeg": TurboJpegLoader,
    "lmdb": LmdbLoader,
    "tfrecords": TFRecordsLoader,
}

顯示圖像：

from argparse import ArgumentParser

import cv2

from loader import (
    CV2Loader,
    LmdbLoader,
    PILLoader,
    TFRecordsLoader,
    TurboJpegLoader,
    methods,
)


def show_image(method, image):
    cv2.imshow(f"{method} image", image)
    k = cv2.waitKey(0) & 0xFF
    if k == 27:  # check ESC pressing
        return True
    else:
        return False


def show_images(loader):
    num_images = len(loader)
    loader = iter(loader)
    for idx in range(num_images):
        image, time = next(loader)
        print_info(image, time)
        stop = show_image(type(loader).__name__, image)
        if stop:
            cv2.destroyAllWindows()
            return


def print_info(image, time):
    print(
        f"Image with {image.shape[0]}x{image.shape[1]} size has been loading for {time} seconds",
    )


def demo(method, path):
    loader = methods[method](path)  # get the image loader
    show_images(loader)


if __name__ == "__main__":
    parser = ArgumentParser()

    parser.add_argument(
        "--path",
        "-p",
        type=str,
        help="path to image, folder of images, lmdb environment path or tfrecords database path",
    )
    parser.add_argument(
        "--method",
        required=True,
        choices=["cv2", "pil", "turbojpeg", "lmdb", "tfrecords"],
        help="Image loading methods to use in benchmark",
    )

    args = parser.parse_args()

    demo(args.method, args.path)

更多細節請參考：

https://github.com/spmallick/learnopencv/tree/master/Efficient-image-loading

https://www.learnopencv.com/efficient-image-loading/

這里就只看結果了：