Well, if a blue splat is (just an example) 93% statistically close to being a swimming pool, there is nothing wrong with upscaling it for many cases where absol

Well, if a blue splat is (just an example) 93% statistically close to being a swimming pool, there is nothing wrong with upscaling it for many cases where absolute accuracy is not required. Larger and better quality datasets will decrease the amount of possible failure modes after a few development / training iterations of this algorithm Of course if outliers / anomalies are very important for persons business or use-case they shouldn't use this feature.