Gdal and Pillow Tiff to Jpeg

Question

I have been using the Pillow code below to convert Tiff to Jpeg, but have recently found that I can use Gdal to also convert Tiff to Jpeg. I am trying to understand what the main difference is between them and if there is a reason why I should be using one over the other? Is there a quality difference? I would prefer to use the gdal code to limit the amount of packages I need in my overarching script.

The purpose is to use the JPEG's within computer vision - tensorflow - pytorch - YOLO - etc.

Thoughts?

Pillow

Image.MAX_IMAGE_PIXELS = None
for infile in os.listdir(tif_path):
    if infile[-3:] == "tif" or infile[-3:] == "bmp":
        outfile = infile[:-3] + "jpeg"
        im = Image.open(tif_path + '/' + infile)
        out = im.convert("RGB")
        out.save(jpeg_dst + outfile, "JPEG", quality=90)

Gdal

    
options_list = [
    '-ot Byte',
    '-of JPEG',
    '-b 1',
    '-scale'
]           

options_string = " ".join(options_list)
    
gdal.Translate(
    'save_image_path.jpg',
    'image_path.tif',
    options=options_string
)

Answer 1

Pillow seems to use the IJG JPEG library https://jpegclub.org/reference/about-ijg/. GDAL may use the same or libtiff-turbo https://gdal.org/drivers/raster/jpeg.html

The GDAL JPEG Driver is built using the Independent JPEG Group’s jpeg library. Also note that the GeoTIFF driver supports tiled TIFF with JPEG compressed tiles. This can be used to apply JPEG compression to datasets that exceed the maximum dimensions of 65,535x65,535 pixels for a single JPEG image.

It is also possible to use the JPEG driver with the libjpeg-turbo, a version of libjpeg, API and ABI compatible with IJG libjpeg-6b, which uses MMX, SSE, and SSE2 SIMD instructions to accelerate baseline JPEG compression/decompression.

Starting with GDAL 3.4, read and write support for JPEG images with 12-bit sample is enabled by default (if JPEG support is also enabled), using GDAL internal libjpeg (based on IJG libjpeg-6b, with additional changes for 12-bit sample support). Support for JPEG with 12-bit sample is independent of whether 8-bit JPEG support is enabled through internal IJG libjpeg-6b or external libjpeg (like libjpeg-turbo)

If you want to be sure that the results are similar use a GDAL version that is build with the IJG JPEG library but the two other options should be compatible as well when it comes to features which are supported by IJG libjpeg-6b.