degrade

Create degraded versions of images for upscale/enhance training. No extra dependencies — pure Pillow.

Usage

datasety degrade --input ./originals --output ./dataset --type random --paired

Degradation Types

Type	Effect at intensity=1.0
lowres	16x bilinear downscale + nearest upscale
oversharpen	Aggressive unsharp mask (percent=1000)
noise	50% Gaussian noise blend
blur	Radius 3px Gaussian blur
jpeg	Quality 5 JPEG compression
motion-blur	Horizontal motion blur
pixelate	16x nearest-neighbor pixelation
color-bands	3-bit posterization
upscale-sim	8x bilinear down + Lanczos up

Options

Option	Description	Default
--input, -i	Input directory	(required*)
--output, -o	Output directory	(required*)
--input-image	Single input image
--output-image	Single output image
--type, -t	Degradation type(s), repeatable	random
--intensity	Global intensity 0.0-1.0	0.5
--intensity-range	Random range MIN-MAX
--chain	Apply multiple types sequentially	false
--num-variants	Variants per input image	1
--paired	Create control/ + target/ subdirs	false
--seed	Random seed	(random)
--output-format	png, jpg, webp	png
--skip-existing	Skip images with existing output	false
--workers	Parallel workers for processing	1
--progress	Show tqdm progress bar	false
--dry-run	Preview without writing files	false

Examples

# Random degradation for upscale training
datasety degrade -i ./originals -o ./dataset --type random --intensity-range 0.2-0.8 --paired

# Chain specific degradations
datasety degrade -i ./images -o ./degraded --type jpeg --type noise --chain --paired

# Multiple variants per image
datasety degrade -i ./images -o ./degraded --type random --num-variants 3 --seed 42