elrandar / linedet-cuda

A massively parallel CUDA implementation of a line detection technique for documents.

kalman-filter-gpu

Compiling

This code requires g++ version 10.

export CXX=/usr/local/bin/g++-10

Before calling cmake.

On a debian computer:

sudo apt install cmake
mkdir build && cd build
cmake ..
make

Usage

Input images are uint8 .PGM ASCII (P2), Range is (0 - 255).

Using ImageMagick, it is possible to convert an image in this format using the following command:

convert <img_path> -colorspace Gray -compress none <out_path>.pgm

The tool is used in the following manner:

./kalman-gpu [mode] <imagepath>

Modes :

• --gpu activate GPU mode
• --batch activate batch CPU mode (CPU mode emulating GPU)
• --sequential activate sequential CPU mode (Original implementation)

The output is a label image in the .PGM format. It is stored in the out.pgm file.
To display the output using python, the following script can be used :

python3 output_to_rgb.py <out_img>