Image Collage Maker

System Requirements

• Python >= 3.5

Install dependencies

pip3 install itchat pillow opencv-contrib-python tqdm scikit-learn umap-learn matplotlib lapjv wurlitzer

Note: You should not install "wurlitzer" if you're on Windows. It helps to visualize the progress of the linear sum assignment by redirecting output from the C++ code to python's stdout, but it does not work on Windows.

How to use

Note: If you already have a set of images to work with, you can skip step 1. The collage maker can be applied to any folder which contains a sufficient amount of images, not limited to your WeChat friends' profile pictures.

1. Use extract_img.py to download profile pictures of your WeChat friends

Download all your friends' profile pictures (--dir specifies the directory to store these images):

python3 extract_img.py --dir img

Or, download the group members' images in a group chat (replace name with the group chat's name and keep the double quotes):

python3 extract_img.py --dir img2 --type chatroom --name "name"

Sometimes the download may fail, especially when the program is running for the first time. In such case, you need to rerun program with an additional --clean flag

python3 extract_img.py --dir img --clean

2. Use make_img.py to make a collage

You can run the GUI to make the collage. Simply type

python3 gui.py

You may also want to try the binaries built by PyInstaller

On MacOS or Linux, after downloading the binary, you may need to add executing permission before launching it. Go to the file's directory and type

chmod +x ./collage-maker-macos-x64

Or, read the following part for the usage of the command-line tool.

Option 1: Sorting

python3 make_img.py --path img --sort pca_bgr --size 100

Use --ratio w h to change the aspect ratio, whose default is 16:9

Example: use --ratio 21 9 to change the aspect ratio to 21:9

Result:

Option 2: Fit a particular image

Option 2.1: Give a fair chance to each image

This fitting option ensures that each image is used for the same amount of times.

python3 make_img.py --path img --collage img/1.png --size 25 --dup 10 --out collage.png

--dup 10 allows each source image to be used 10 times. Increase that number if you don't have enough source images or you want a better fitting result. Note that a large number of images may result in long computational time.

Original	Fitting Result

Option 2.2: Best fit

This fitting option just selects the best subset of images you provided to approximate your destination images. Each image in that subset will be used for an arbitrary number of times.

Add --uneven flag to enable this option. You can also specify --max_width to change the width of the grid. The height will be automatically calculated based on the max_width provided. Generally, a larger grid will give a better result. The default value is 80.

python3 make_img.py --path img --out collage-best-fit.png --collage img/1.png --size 25 --uneven

Original	Fitting Result

Option 2.3 Display salient object only

This fitting option only fits the source images with the pixels of the destination image that constitute salient objects.

Add --salient flag to enable this option. You can still specify whether each image is used for the same amount of times with the --uneven flag.

Use --lower_thresh to specify the threshold for object detection. The threshold ranges from 0 to 225; a higher threshold would lead to less object area. The default threshold is 75. If you choose to use each image for the same amount of time, the threshold may have to change so that the number of source images and the number of pixels in the destination can converge. You may use --lower_thresh -1 to enable adaptive thresholding (new in v.2.1).

Use --background to specify the background color for the collage. The color space for the background option is RGB. The default background color is white, i.e. (255, 255, 255).

python3 make_img.py --path img --out collage-best-fit.png --collage img/1.png --dup 16 --salient

python3 make_img.py --path img --out collage-best-fit.png --collage img/1.png --size 25 --salient --uneven

Original	Uneven-Fitting Result	Even-Fitting Result

Other options

Use --sigmato specify the multiplier of the identity matrix used as the covariance matrix of the bivariate normal distribution from which the weights of each pixel are sampled. A positive sigma implies a higher weight for the pixels in the middle of the image. A negative sigma indicates higher weight for the pixels close to the edge of the image. The greater the absolute value of the sigma, the greater the difference of weights between the pixels in the middle and the pixels on the edge will be.

Use --exp to traverse every possible sigmas and colorspace. If you use the option --salient, use --exp to traverse thresholds from 40 to 190.

Use python3 make_img.py --help to get other optional arguments.

Mechanism

A brief description of the mechanism is available here

Credits (Names in alphabetical order)

Hanzhi Zhou (hanzhi713): Main algorithm and GUI implementation

Kaiying Shan (kaiyingshan): Saliency idea and implementation

Xinyue Lin: Idea for the "Uneven distribution (best-fit)"

Yufeng Chi (T-K) : Initial Idea, crawler