aimerykong / predictive-filter-flow

Predictive Filter Flow for fully/self-supervised learning on various vision tasks
https://www.ics.uci.edu/~skong2/mgpff.html
140 stars 24 forks source link

some image is loaded upside down #4

Open zlw21gxy opened 5 years ago

zlw21gxy commented 5 years ago

image

zlw21gxy commented 5 years ago

I wonder why we need do it?Flip image is some kind like data argument? if self.set_name=='train' and np.random.random(1)>0.5: image1 = image1.transpose(PIL.Image.FLIP_TOP_BOTTOM) image2 = image2.transpose(PIL.Image.FLIP_TOP_BOTTOM) if self.set_name=='train' and np.random.random(1)>0.5: image1 = image1.transpose(PIL.Image.FLIP_LEFT_RIGHT) image2 = image2.transpose(PIL.Image.FLIP_LEFT_RIGHT)

aimerykong commented 5 years ago

Yes. It is for augmenting training data. As mgPFF is a low-level vision method, meaning that it does not need to understand high-level semantics, such flip and flop augmentation can be applied here. But it is optional when you have enough training data.

On Sun, Apr 28, 2019 at 7:53 PM zlw21gxy notifications@github.com wrote:

I wonder why we need do it?Flip image is some kind like data argument? if self.set_name=='train' and np.random.random(1)>0.5: image1 = image1.transpose(PIL.Image.FLIP_TOP_BOTTOM) image2 = image2.transpose(PIL.Image.FLIP_TOP_BOTTOM) if self.set_name=='train' and np.random.random(1)>0.5: image1 = image1.transpose(PIL.Image.FLIP_LEFT_RIGHT) image2 = image2.transpose(PIL.Image.FLIP_LEFT_RIGHT)

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/aimerykong/predictive-filter-flow/issues/4#issuecomment-487439941, or mute the thread https://github.com/notifications/unsubscribe-auth/ABRJSJHCDKVKJVASUUXUDWDPSZPKRANCNFSM4HI76GKA .