original images: 10870 x 10870
result images after: 2047 x 2047
I chose nothing but auto-orient in augmentation.
Is 2047 the maximum resolution supported by roboflow?
Hi @gilm - I just spoke with the team and there’s functionality in our backend to automatically resize images down to a maximum of 2048 x 2048, otherwise it crashes memory-wise.
The image transformation operations (dataset generation steps) become very slow for images of greater sizes. Dataset generation essentially creates new copies of each individual image.
Images of a size of 10870 x 10870 would greatly slow things down for all other users if they were accepted for dataset generation at that size.
Are you working with aerial or satellite imagery, and training a model to make detections in them? Or are these high-resolution images for another task/source? You’ll want to be using Tiling (a Preprocessing step) during dataset generation, as images of over 10k pixels by 10k pixels would give you a very slow model training time; are likely be too big to fit into memory; and give you very slow inference speeds.
