Skip to content

Latest commit

 

History

History
46 lines (30 loc) · 1.6 KB

README.md

File metadata and controls

46 lines (30 loc) · 1.6 KB

EPIS

An implementation of the Edge-Preserving Image Smoothing (EPIS) algorithm discussed in this paper.

Our goal is to measure the impact of L1 piece-wise flattening with edge-preserving on image compression for various image formats.

Showcase

EPIS transforms an image (left) into a flattened image (right).

EPIS showcase

Usage

python epis.py [filepath to image]

TODO

  • Speedup compute_pairs() by generating the matrices pair1 and pair2 instead building them iteratively.

Requirements

Libraries used and their respective versions. The minimum versions for each library are unknown. A detailed list can be found in library_versions.txt.

  • Python 3.10.0
  • Numpy 1.24.2
  • CuPy 12.0.0
  • CuSPARSE 0.4.0
  • CUDA 12.1

Hardware

The hardware used for test_data are:

  • AMD Ryzen 5 5600x 6-Core 12-Thread
  • 8GB Nvidia RTX 3070
  • 32 GB DDR4-3600 RAM

EPIS relies heavily on GPU computation, specifically sparse matrix multiplication and sparse matrix solving. EPIS uses incredibly large sparse matrixes which necessitates a large pool of VRAM. This is the limiting factor when flattening images. Downscaling images is required depending on the hardware used.

A GPU with 8GB can flatten images with dimensions of about 180x180 pixels. An example hardware usage analysis on architecture_118_180.png is shown below.

Hardware Usage