How would you go about training this data on a larger/more diverse data set? I've looked into some open source data from SpaceNet, and was wondering if I could use that data to deal with the overfitting issue (https://spacenetchallenge.github.io/datasets/spacenetBuildings-V1summary.html). I've noticed that these images are not 300x300 pixel though, and wanted to know if the data would still work for training. Doing image manipulation would be simple, but it would be tough to convert all of the pixel-based labels based off different dimensions.
How would you go about training this data on a larger/more diverse data set? I've looked into some open source data from SpaceNet, and was wondering if I could use that data to deal with the overfitting issue (https://spacenetchallenge.github.io/datasets/spacenetBuildings-V1summary.html). I've noticed that these images are not 300x300 pixel though, and wanted to know if the data would still work for training. Doing image manipulation would be simple, but it would be tough to convert all of the pixel-based labels based off different dimensions.