Batch-processes images (rotate, blur, brightness/contrast/saturation) for ML training data.
pip install -r preprocess/requirements.txt
python augment_images.py <files-or-folders> -o <output_dir> [options]
| Flag | Description |
|---|---|
-o, --output |
Output folder (default: preprocess/image_output/augmented_images) |
--rotate DEG |
Rotation angle in degrees |
--blur RADIUS |
Gaussian blur radius |
--brightness F |
Brightness factor (1.0 = unchanged) |
--contrast F |
Contrast factor (1.0 = unchanged) |
--saturation F |
Saturation factor (1.0 = unchanged) |
--random |
Randomize each param within the given value as a range instead of applying it as fixed |
--copies N |
Number of augmented copies per source image (default: 1) |
--limit N |
Max number of source files to process |
--max-dim N |
Cap the longest side of output images to N pixels |
Single image, fixed transforms:
python3 preprocess/augment_images/augment_images.py photo.jpg --rotate 15 --blur 2
Folder, randomized augmentation, 3 variants each, capped resolution:
python3 preprocess/augment_images/augment_images.py /path/to/folder --brightness 0.2 --contrast 0.3 --saturation 0.3 --random --copies 3 --max-dim 1024
Process only the first 50 files in a folder:
python3 preprocess/augment_images/augment_images.py /path/to/folder --rotate 10 --limit 50
- Supported input formats: jpg, jpeg, png, bmp, webp
- Outputs keep the original file extension
--randomranges:--rotate 15→ random angle between -15 and 15;--brightness 0.2→ random factor between 0.8 and 1.2