/CNN-Learns-Without-Boundaries

Do cnn requires edes/boundaries/pixelvariation to learn? Can CNNs learn without explicit pixel variation(information not explicilty represented by pixels in an image)

Primary LanguagePython

CNN-Learns-Without-Boundaries

Do CNNs require edes/boundaries/pixelvariation to learn? Can CNNs learn without explicit pixel variation(information not explicilty represented by pixels in an image)

Problem Statement : Given two points,can a cnn learn a function that draws a circle with the points as the ends of its diameter.

alt text

Input Image

alt text

GroundTruth Mask

alt text

Model Output Mask

References

model.py, loss.py, helper.py are taken from #https://github.com/usuyama/pytorch-unet