nvjpeg2k-python

NVJPEG2K Python binding, this is mainly used for the Kaggle competition RSNA Screening Mammography Breast Cancer Detection so the functionality may not completed, for now it only support decode and lack of test.

See this kaggle notebook that use this library to boost the image loading speed.

Steps compile

Requirements

Linux
cmake
CUDA 11.x
Python 3.6 or later
libnvjpeg_2k, you can download it from here, be sure to download the nvJPEG2000.

Build

# get the source code
git clone --recursive https://github.com/louis-she/nvjpeg2k-python.git
cd nvjpeg2k-python

# build the extension
mkdir build
cd build

# **ABSOLUTE PATH** of libnvjpeg2k path, modified this in your own case
PATH_OF_THE_LIBNVJPEG_2K="/....../libnvjpeg_2k-linux-x86_64-0.6.0.28-archive"

# can get rid of the `-DCMAKE_BUILD_TYPE=Debug` option
cmake .. \
  -DCMAKE_BUILD_TYPE=Debug \
  -DNVJPEG2K_PATH=${PATH_OF_THE_LIBNVJPEG_2K} \
  -DNVJPEG2K_LIB=${PATH_OF_THE_LIBNVJPEG_2K}/lib/libnvjpeg2k_static.a

make

Usage

import nvjpeg2k
from pathlib import Path

decoder = nvjpeg2k.Decoder()
res = decoder.decode(Path("../assets/demo.dcm.jp2").read_bytes())
print(res.shape)