Invisibility Cloak OpenCV with Deep Learning

Invisibility Cloak Project

Computer Vision with OpenCV

Graduation Project in 2021 Fall Semester @UNIST

For this project, FGVC(Flow-edge Guided Video Completion) deep learning model was used for object removal.

Result

✔ Youtube | Presentation (Kor) | Report (Kor)

Environment

Package	Version
anaconda (x64)	4.10.3
cuda	10.2.89
matplotlib	3.4.3
numpy	1.21.4
opencv	4.5.4
os	Windows 10
pip	21.0.1
python	3.8.12
pytorch	1.6.0
scipy	1.6.2

Usage

Install

$ conda create -n (name)
$ conda activate (name)
$ conda install pytorch==1.6.0 torchvision==0.7.0 cudatoolkit=10.2 -c pytorch
$ pip install matplotlib scipy opencv imageio imageio-ffmpeg scikit-image imutils

Run Project

• Download and unzip weight.zip into the modules.
• Prepare video sequences dataset of color and mask for project. (Data Samples : tennis | f250)
• Run project

# Remove __pycache__ if you want.
$ find . | grep -E "(__pycache__|\.pyc|\.pyo$)" | xargs rm -rf

# Run video inpainting.
$ python run_inpainting.py \
> --path      './data/tennis_color' \
> --path_mask './data/tennis_mask' \
> --outroot   './data/tennis_result' \
> --merge \
> --run

Update History

v1.3

• Add to remove object with FGVC
• Set the main cloak color with RED
• Update cloak mask and noise.
• Rename ftn -> image_stack.

v1.2

• Add to save output video.
• Add color selection mode : RED || GREEN
• Get object removal result by FGVC.

v1.1

• Add specific noise filtering conditions.
• Modify detected color : RED -> GREEN

v1.0

• Project first commit.
• Test module ftn for showing image stack view.
• Test HSV Detector using track bar.
• Test Invisibility Cloak Demo.

---

Updated : 2021-11-28 00:19