Comparison of Deep Block and Simple Tiling

In this project, we trained an image segmentation model to segment rice paddy areas in 5600x4600 aerial photos.
For training, we divided each aerial photo into 7 rows and 5 columns.
During this process, Deep Block automatically performs annotation and image splitting.
After training the model for 10 epochs, we ran inference using a simple image splitting method and our patented tiling method and compared the results.
Analyze your images using Deep Block's unique tiling options.

Visit https://www.deepblock.net/blog/deep-block-algorithm-for-large-scale-images
for more information