Develop a docker image(thumbnail generator)

As described in Pipelines and dockers, there are four types of docker images, namely the Image preprocessor, featurizer, thumbnail generator and attribute generator. This article describes the general process for developing a docker image using a thumbnail generator as an example. The rest of the articles detail the interface specification for other types of dockers and the fields required for registering other types of dockers.

Interface specification

Sample program

Based on the above specification, a sample implementation(in Python) is provided below. This program does the following.

1. Reads files from /input/i1
2. For each file, extracts the filename from the path and write out a thumbnail with the exact same filename preserving the path relative to /input/i1.

import os

from PIL import Image, ImageFile, UnidentifiedImageError

OUTPUT_WIDTH = 192
OUTPUT_HEIGHT = 108
ImageFile.LOAD_TRUNCATED_IMAGES = True
SOURCE_DIR = '/input/i1'
DEST_DIR = '/output/o1'


def get_input_files():
    file_list = []
    for root, _, files in os.walk(SOURCE_DIR):
        for filename in files:
            file_list.append(os.path.join(root, filename))
    return file_list


def generate_thumbnail_and_save(fname, width, height):
    try:
        with Image.open(fname) as img:
            img.thumbnail((width, height), resample=Image.BICUBIC)
            # Extract filename and save thumbnail with same filename
            dest_path = os.path.join(DEST_DIR, fname[len(SOURCE_DIR) + 1:])
            path_dir = os.path.dirname(dest_path)
            if not os.path.exists(path_dir):
                os.makedirs(path_dir)
            img.convert('RGB').save(dest_path, format="JPEG", quality=95)
    except UnidentifiedImageError as ex:
        print("Unknown Image format {}".format(fname))
        raise ex
    except Exception as ex:
        print("Failed to generate thumbnail for {}".format(fname))
        raise ex


def main():
    input_files = get_input_files()
    # Throw error if input is empty
    if len(input_files) == 0:
        print("Input file list to create thumbnail is empty")
        raise ValueError("Input file list to create thumbnail is empty")

    for f in input_files:
        generate_thumbnail_and_save(f, OUTPUT_WIDTH, OUTPUT_HEIGHT)

if __name__ == "__main__":
    if not os.path.exists(DEST_DIR):
        os.makedirs(DEST_DIR)
    main()

Packaging into a docker

Below is a sample Dockerfile to build docker with the above Python program and necessary dependencies. The Dockerfile assumes the above Python program is stored as cmds/thumbnail.py relative to where this build is being done and copies the file into /opt/akridata/thumbnail.py within the docker.

FROM python:3.8-slim-buster

RUN python3.8 -m pip install --upgrade pip==21.2.4 pillow==9.0.1
RUN mkdir /opt/akridata/
COPY cmds/thumbnail.py /opt/akridata/thumbnail.py

With the above stored as Dockerfile, the docker image can be built using the following command.

docker build -t thumbnail_generator:latest .

The docker can be tested using a sample command as below. The test assumes one or more input images are stored in /tmp/i1. The -v option mounts /tmp/i1 at /input/i1 within the docker and /tmp/o1 at /output/o1 within the docker. At the end of a successful run, the output thumbnails are expected to be present in /tmp/o1 with the same file names as the files in /tmp/i1.

docker run -v /tmp/i1:/input/i1 -v /tmp/o1:/output/o1 thumbnail_generator:latest python3 /opt/akridata/thumbnail.py

Upload docker to a registry(docker repository)

The docker image must be uploaded to the registry using the docker push command. Please refer to Publish-docker-image for more details.