FriendBlend

FriendBlend is an method to perform intelligent image composition, enabling two people to combine two photos of them with the same background into a single one, as if both people were present in the scene together.

This project was built as part of the Digital Image Processing Course at IIIT Hyderabad.

Team Information

Name	Roll Number	Branch
Ahish Deshpande	2018102022	ECE
Pranav Kirsur	2018101070	CSE
Pranav Tadimeti	2018101055	CSE
Yoogottam Khandelwal	2018101019	CSE

Project code layout

.
├── app              # mobile app (abandoned after mid-eval)
├── documents        # presentation related files
├── images           # contains images we used for running the algorithm
├── Makefile          # helper commands for quickly building and running the project
├── misc             # random experiments we did while implementing the paper
├── README.md        # you are here
└── src              # the project implementation

The source code has been ordered in a logical manner instead of just putting everything into a single file.

src
├── Dockerfile
├── friendblend                      # main module
│   ├── global_vars.py               # contains module level per-process global variables
│   ├── helpers.py                   # basic helper functions (logging, timing, etc.)
│   ├── main.py                      # main code, the algorithm
│   ├── processing                   # contains code related to individual steps in the pipeline
│   │   ├── alpha_blending.py
│   │   ├── color_correction.py
│   │   ├── face_body_detection.py
│   │   ├── grab_cut.py
│   │   ├── helpers.py
│   │   ├── __init__.py
│   │   ├── keypoint.py
└── requirements.txt                 # pip install this

Code Documentation

Link to documentation: https://yoogottamk.github.io/Intelligent-Image-Composition/

How to run?

We assume that you will be running this on a linux system which has the GNU standard C Library (most of the normal machines have this).

Prerequisites

• python3.8
• pip (corresponding to python3.8)
  OR
• docker

The normal way

Since you most probably don't want to mess up your system's python installation, it's best if you create a python virtual environment.

project-appy-fizz$ pip install -U virtualenv
project-appy-fizz$ virtualenv venv -p python3.8
project-appy-fizz$ source venv/bin/activate

Install the requirements

project-appy-fizz$ pip install -r src/requirements.txt

Now, go inside the src directory.

project-appy-fizz$ cd src

Here, source the .env file which sets the environment variable $PYTHONPATH

project-appy-fizz/src$ source .env

Now, run main.py from inside the same directory

project-appy-fizz/src$ python friendblend/main.py

This runs it with the default fallback images. If you want to run this on your own images, please put them inside the images directory and

project-appy-fizz/src$ python friendblend/main.py F1.ext F2.ext

where files F1.ext and F2.ext are present in the images directory inside our repository.

NOTE F1.ext is just an example, naming myfile.png also works.

The docker way

If you have docker installed, it's very easy to run our code.

Ensure you are at repository root.

To build the docker image,

project-appy-fizz$ make build

To run the docker image,

project-appy-fizz$ make run

Note that in order to run it with custom images, you'll have to run the actual docker run command instead of using the Makefile.

project-appy-fizz$ docker run --rm -it -v ${PWD}/images:/images friendblend F1.ext F2.ext

where files F1.ext and F2.ext images are present in the images directory inside our repository.

The outputs of each intermediate step of the friendblend algorithm, along with the final output, are stored in images/outputs/ directory.