Here are some random images I colorized tonight using my unreleased DeOldify model. 3/4 are from Reddit's r/ColorizationRequests. Enjoy!pic.twitter.com/o7rZY0IRNF
-
I like to create an evolution timeline with such models. At the end of the day, I get to watch a cool 120fps/60fps movie of how the network learns to do the job. Plus it also lets me check in on it during training time
1/ Tensorboard allows scrubbing of image changes as they train over time which is what I use. But the selection of which frames get used is super goofy IMHO and doesn’t make much intuitive sense. A linear progression via a high FPS movie would be great.
-
-
2/ I can tell you this much though- being able to see -changes- like that has been key to catching issues, and catching them early. Some issues really are subtle and are best revealed via animation.
-
Yeah so just pick your poison(read: most troublesome photos) and make that animation. It was Soo cool when it first learnt to color sky blue and vegetation as green
End of conversation
New conversation -
-
-
I just make a giant grid in matplolib with 9 pictures let's say and make a 3*3 matrix and then just color that every n-steps or so
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.