Drop images and optional labels
Images in any browser format (JPEG, PNG, WebP). YOLO txt files with matching names are picked up automatically.
Datasets
Batch Resize & Augment
Prepare images at the size a model trains on, and generate simple augmentations, without a Python environment. Choose a target size and fit mode, pick augmentations, and download a zip. If YOLO label files are dropped alongside, boxes are transformed with the images so flips and letterboxing keep labels correct.
Loading the tool…
How it works
Choose size and augmentations
Fit modes: stretch, letterbox (pad), crop to fill. Augmentations: horizontal flip, vertical flip, 90-degree rotations, brightness and contrast ranges, greyscale, blur.
Download
A zip with the processed images (and labels) plus a manifest of what was applied to each file.
Formats, limits and the numbers that matter
Label transforms
Flips and rotations are exact. Letterboxing rescales and offsets boxes. Crop-to-fill drops boxes that fall outside and clips those on the edge, which is reported.
Quality
JPEG output quality is settable; PNG keeps transparency. Files are re-encoded, so use PNG for masks.
Why augment offline?
Most trainers augment on the fly, which is better. Offline augmentation is for small datasets, for frameworks without augmentation, and for producing a fixed evaluation set.
Common questions
What size for YOLO?
640 is the default and fine for most; go to 1024 or 1280 when boxes are small relative to the image. Letterbox rather than stretch.
Does it handle EXIF orientation?
Yes; images are drawn upright before resizing, and labels are assumed to be for the upright image.
How many at once?
Limited by browser memory; a thousand 12-megapixel photos is comfortable on a laptop.