Deep Object Removal

Image completion is a challenging problem because it requires a high-level recognition of scenes. This project tries to achieve object removal from images and get the base image reconstructed based on surrounding colours and objects using conditional GANs.

Overview

This project is an implementation of cGANs discussed in the paper for [General Image Completion]
The models are tweaked a little and implemented to remove objects from images and reconstruct the image without the object.

Example Usage

Hot Keys

[Esc]: To quit the windowed application.
[f]: To filter out the masked object.
[n]: To go to the next image.
[r]: To refresh and undo all the masking in the current image.

Description

Files

images/

The folder that contains the images to be used in the project. Currently the project requires images of dimensions 400 x 400 which can be changed in the main.py file.

model/

This folder contains the pretrained model that is trained on mscoco dataset and the model definition file which is written in tensorflow.

main.py

The main file to run the program. The code runs as an OpenCV windowed application.

requirements.txt

The requirements file for the project

Installation

To install the dependencies type

sudo pip3 install -r requirements.txt

Run

To run the application type

python3 main.py

This will run the demo as an OpenCV application

Dependencies

The project requires the following packages:

OpenCV and OpenCV_python 3.3.0.10
Tensorflow 1.10.1
Numpy 1.13.3