Hey, two questions: Do you have to create the ground thruths by hand? And is upsampling integrated to pytorch or tensorflow? Because if you have to do it yourself it would take a lot of time to run, right?
Transpose 2D Convolution layers are used for upsampling. They're just layers you add into the model. Here I used an annotation tool to create ground truths. Typically, you would use an annotated dataset and train the model but the way I set this series up was to let you annotate your own data. It takes a while. I think there are faster tools out there for semantic segmentation.
"Classification models are really boring", hahahah, true!
Your content is great, Seth! Thank you!
can you make video on pytorch implementation
How can we see intermediate things using tensorboard?
is this unsupervised learning method or supervised learning like U-Net?
Hey, two questions: Do you have to create the ground thruths by hand?
And is upsampling integrated to pytorch or tensorflow? Because if you have to do it yourself it would take a lot of time to run, right?
Transpose 2D Convolution layers are used for upsampling. They're just layers you add into the model. Here I used an annotation tool to create ground truths. Typically, you would use an annotated dataset and train the model but the way I set this series up was to let you annotate your own data. It takes a while. I think there are faster tools out there for semantic segmentation.
👍🏻