benchmark_compress.py

import io
import gzip
import bz2
import lzma
import numpy as np
from utils.torch.modules import ImageNet
import os
from PIL.PngImagePlugin import getchunks

from torchvision import datasets, transforms
import PIL.Image as pimg

# code that applies benchmark compressors on the three datasets (MNIST, CIFAR-10 and ImageNet)
# heavily based on benchmark_compressors.py from https://github.com/bits-back/bits-back

# seed (for reproducibility MUST be the same for the whole project on the same machine)
np.random.seed(100)

# method to extract maximum amount of pixel-blocks of certain size from certain image
def extract_blocks(arr, block_size=(32, 32)):
    nrows, ncols = block_size
    h, w, c = arr.shape
    if h % nrows != 0:
        h -= h % nrows
        arr = arr[:h]
    if w % ncols != 0:
        w -= w % ncols
        arr = arr[:,:w]
    return (arr.reshape(h//nrows, nrows, -1, ncols, c)
               .swapaxes(1,2)
               .reshape(-1, nrows, ncols, c)), h, w

# method to reconstruct image from the extracted pixel-blocks
# note: this returns the original images being cropped to multiples of 32 pixels on each side
def unextract_blocks(arr, h, w):
    n, nrows, ncols, c = arr.shape
    return (arr.reshape(h//nrows, -1, nrows, ncols, c)
               .swapaxes(1,2)
               .reshape(h, w, c))
class ToInt:
    def __call__(self, pic):
        return pic * 255

def mnist(exp):
    transform_ops = transforms.Compose([transforms.ToTensor(), ToInt()])
    mnist = datasets.MNIST(root="model/data/mnist", train=False, transform=transform_ops, download=True)
    return mnist.test_data.numpy()[np.random.choice(len(mnist.test_data), size=(100, 100), replace=False)[exp]]

def cifar(exp):
    transform_ops = transforms.Compose([transforms.ToTensor(), ToInt()])
    cifar = datasets.CIFAR10(root="model/data/cifar", train=False, transform=transform_ops, download=True)
    return cifar.test_data[np.random.choice(len(cifar.test_data), size=(100, 100), replace=False)[exp]]

def imagenet(exp):
    transform_ops = transforms.Compose([transforms.ToTensor(), ToInt()])
    data = ImageNet(root='model/data/imagenet/test', file='test.npy', transform=transform_ops)
    if not os.path.exists("bitstreams/imagenet/indices"):
        randindices = np.random.choice(len(data.dataset), size=(100, 100), replace=False)
        np.save("bitstreams/imagenet/indices", randindices)
    else:
        randindices = np.load("bitstreams/imagenet/indices")
    return data.dataset[randindices[exp]]

def gzip_compress(images):
    images = np.packbits(images) if images.dtype is np.dtype(bool) else images
    assert images.dtype == np.dtype('uint8')
    return gzip.compress(images.tobytes())

def bz2_compress(images):
    images = np.packbits(images) if images.dtype is np.dtype(bool) else images
    assert images.dtype == np.dtype('uint8')
    return bz2.compress(images.tobytes())

def lzma_compress(images):
    images = np.packbits(images) if images.dtype is np.dtype(bool) else images
    assert images.dtype == np.dtype('uint8')
    return lzma.compress(images.tobytes())

def pimg_compress(format='PNG', **params):
    def compress_fun(images):
        compressed_data = bytearray()
        for n, image in enumerate(images):
            img = pimg.fromarray(image)
            if format == 'PNG':
                for chunk_type, chunk_data, crc in getchunks(img, optimize=True):
                    if chunk_type == b'IDAT':
                        compressed_data.extend(chunk_data)
            else:
                img_bytes = io.BytesIO()
                img.save(img_bytes, format=format, **params)
                compressed_data.extend(img_bytes.getvalue())
        return compressed_data
    return compress_fun

def gz_and_pimg(images, format='PNG', **params):
    pimg_compressed_data = pimg_compress(images, format, **params)
    return gzip.compress(pimg_compressed_data)

def bench_compressor(compress_fun, images):
    byts = compress_fun(images)
    n_bits = len(byts) * 8
    bitsperdim = n_bits / np.size(images)
    return bitsperdim

if __name__ == "__main__":
    gzip_list = []
    bz2_list = []
    lzma_list = []
    png_list = []
    webp_list = []
    # MNIST
    print(f"Compressing MNIST test set")
    for exp in range(100):
        images = mnist(exp)
        gzip_list.append(bench_compressor(gzip_compress, images))
        bz2_list.append(bench_compressor(bz2_compress, images))
        lzma_list.append(bench_compressor(lzma_compress, images))
        png_list.append(bench_compressor(
            pimg_compress("PNG", optimize=True), images))
        webp_list.append(bench_compressor(
            pimg_compress('WebP', lossless=True, quality=100), images))
    print(f"gzip: {np.mean(gzip_list):.2f} bits/dim")
    print(f"bz2: {np.mean(bz2_list):.2f} bits/dim")
    print(f"lzma: {np.mean(lzma_list):.2f} bits/dim")
    print(f"png: {np.mean(png_list):.2f} bits/dim")
    print(f"webp: {np.mean(webp_list):.2f} bits/dim")
    print("")

    gzip_list = []
    bz2_list = []
    lzma_list = []
    png_list = []
    webp_list = []
    print(f"Compressing CIFAR-10 test set")
    for exp in range(100):
        # CIFAR-10
        images = cifar(exp)
        # MNIST
        gzip_list.append(bench_compressor(gzip_compress, images))
        bz2_list.append(bench_compressor(bz2_compress, images))
        lzma_list.append(bench_compressor(lzma_compress, images))
        png_list.append(bench_compressor(
            pimg_compress("PNG", optimize=True), images))
        webp_list.append(bench_compressor(
            pimg_compress('WebP', lossless=True, quality=100), images))
    print(f"gzip: {np.mean(gzip_list):.2f} bits/dim")
    print(f"bz2: {np.mean(bz2_list):.2f} bits/dim")
    print(f"lzma: {np.mean(lzma_list):.2f} bits/dim")
    print(f"png: {np.mean(png_list):.2f} bits/dim")
    print(f"webp: {np.mean(webp_list):.2f} bits/dim")
    print("")

    gzip_list = []
    bz2_list = []
    lzma_list = []
    png_list = []
    webp_list = []
    print(f"Compressing 10000 images from ImageNet test set")
    for exp in range(100):
        # ImageNet
        images = imagenet(exp)
        gzip_list.append(bench_compressor(gzip_compress, images))
        bz2_list.append(bench_compressor(bz2_compress, images))
        lzma_list.append(bench_compressor(lzma_compress, images))
        png_list.append(bench_compressor(
            pimg_compress("PNG", optimize=True), images))
        webp_list.append(bench_compressor(
            pimg_compress('WebP', lossless=True, quality=100), images))
    print(f"gzip: {np.mean(gzip_list):.2f} bits/dim")
    print(f"bz2: {np.mean(bz2_list):.2f} bits/dim")
    print(f"lzma: {np.mean(lzma_list):.2f} bits/dim")
    print(f"png: {np.mean(png_list):.2f} bits/dim")
    print(f"webp: {np.mean(webp_list):.2f} bits/dim")
    print("")