realsr-ncnn-vulkan
TLDR
Upscale image
$ realsr-ncnn-vulkan -i [input.jpg] -o [output.png]
Specify scale factor$ realsr-ncnn-vulkan -i [input.jpg] -o [output.png] -s [4]
Use specific model$ realsr-ncnn-vulkan -i [input.jpg] -o [output.png] -m [models-DF2K]
Specify GPU$ realsr-ncnn-vulkan -i [input.jpg] -o [output.png] -g [0]
SYNOPSIS
realsr-ncnn-vulkan [options] -i input -o output
DESCRIPTION
realsr-ncnn-vulkan is a Real-ESRGAN/Real-SR image upscaler using the ncnn neural network framework with Vulkan GPU acceleration. It produces high-quality upscaled images.
PARAMETERS
-i path
Input image file.-o path
Output image file.-s scale
Upscale ratio (4 default).-m path
Model directory.-g gpu
GPU device ID (-1 for CPU).-t size
Tile size.-j threads
Processing threads.
EXAMPLES
$ # Basic 4x upscale
realsr-ncnn-vulkan -i photo.jpg -o photo_4x.png
# 2x scale
realsr-ncnn-vulkan -i photo.jpg -o photo_2x.png -s 2
# Use specific model
realsr-ncnn-vulkan -i anime.png -o anime_up.png -m models-DF2K_JPEG
# CPU only
realsr-ncnn-vulkan -i input.jpg -o output.png -g -1
# Batch process (use loop)
for f in *.jpg; do
realsr-ncnn-vulkan -i "$f" -o "upscaled_$f"
done
realsr-ncnn-vulkan -i photo.jpg -o photo_4x.png
# 2x scale
realsr-ncnn-vulkan -i photo.jpg -o photo_2x.png -s 2
# Use specific model
realsr-ncnn-vulkan -i anime.png -o anime_up.png -m models-DF2K_JPEG
# CPU only
realsr-ncnn-vulkan -i input.jpg -o output.png -g -1
# Batch process (use loop)
for f in *.jpg; do
realsr-ncnn-vulkan -i "$f" -o "upscaled_$f"
done
MODELS
$ models-DF2K - General purpose
models-DF2K_JPEG - JPEG artifact reduction
models-DF2K_JPEG - JPEG artifact reduction
CAVEATS
Requires Vulkan-capable GPU. Memory-intensive for large images. Processing can be slow.
HISTORY
realsr-ncnn-vulkan is based on Real-ESRGAN research, ported to ncnn by nihui for efficient cross-platform inference.
SEE ALSO
waifu2x-ncnn-vulkan(1), esrgan(1), imagemagick(1)
