In this video, we will build an AI model to extract buildings in aerial photos using DEEP BLOCK, and based on this model, we will check the model's inference results on validation data. (1.225GP image)
In this project, we used the aerial image dataset published by INRIA.
The resolution of the training image is 5000x5000 pixels, and we will divide it into 5x5 to train the model.
Click the DIVIDE & TRAIN button.
After some time, the model will start training and a graph will be displayed showing the training progress.
Training the model on the training data takes approximately 20 minutes.
Over time, training is complete.
Now, we will analyze the super-resolution images with the trained model.
Click the PREDICT button and start running the model with the 35x35 division option for our validation image.
After a certain period of time, the inference of the model is completed.
We stitched together several validation images to create a 35000x35000 pixel image.
A single image is several gigabytes in size.
There are more than 30,000 buildings in the image.
The remote sensing images we have to deal in the real world are very large, and powerful software technologies are needed to analyze these images.
Get started with DEEP BLOCK, a very easy and seamless remote sensing image analysis AI model development tool now!