Fantastic Filter CLI

A CLI tool for Fantastic Filter written in C.

Usage

Enter fantastic-Filter-cli --help for helps.

TL;DR

Run Benchmark

fantastic-Filter-cli -b[=LEVEL] # where LEVEL is 0-4

Note: Level 0 is for testing, which might be TOO stressless for GPU, but enough for CPU.

Process An Image

fantastic-Filter-cli -o <OUT_DIR> [-w=MAX_HEIGHT] [-h=MAX_HEIGHT] <input_files...>

MAX_HEIGHT & MAX_WIDTH depends on how much (V)RAM you have. For example, an RTX 2080ti with 11GB VRAM can process images larger than 3840x2160, but 1920x1080 is too much for a GTX 1050 with 2GB VRAM. Try to reduce the size if you get OOM error.

Pre-requirements

• libtensorflow (tested: 2.3 2.4)
• gcc or clang (set $CC)
• make

Optional

• cuda & cudnn (tested: cuda-10.1 & cudnn-7)

Build & Install

make

After make, ./bin/fantastic-filter-cli will be generated. Move it to somewhere if you want.

Download FF_Model.tar.gz from releases, extract it, and set the environment variable FF_MODEL_PATH to the directory.

Docker

The basic idea is to mount the directory where the image you want to process to docker.

Here's an example:

docker run --gpus all -v <dir_on_host>:/myDir -it <image_name> fantastic-filter-cli tmp -o /myDir <filename>

TODO

Build an app image.