Day 11
There wasn’t a meeting today, so I went straight to my lab.
This morning I got another paper (on DualGAN) and my final project: coding a dual learning GAN to translate thermal images to RGB as well as comparing how the results differ when using unpaired/unlabeled inputs versus paired/labeled inputs. Paired/labeled data will typically result in higher accuracy, however in reality, obtaining large amounts of it can be time-consuming, difficult, or impossible. So unpaired/unlabeled data is preferable in many applications as long as the accuracy isn’t too much lower. Besides the usual adversarial loss, a dual learning GAN gets around the problem of unpaired/unlabeled inputs by using a cycle-consistency loss or, as the DualGAN paper called it, a reconstruction loss. You can think of it as translating something forward and back and comparing how close the result is to the original. So there are two pairs of generators and discriminators at work rather than just one. I think that after I compare the unpaired and paired input settings, I’ll be comparing my code to another one.
I’m now in the reading room as the computers in there have a greater capacity than the ones functioning in the lab. I'm working on the model with the unpaired inputs first, and then I'll alter it for the paired inputs. Right now I’m coding the generators. They have a U-Net structure, which means that they have skip connections in the layers that allow low-level information to move directly across the network instead of all filtering through the bottleneck in the middle. It’s really interesting to see how this project is building off the papers I read during the first week, since they had information on U-Nets, cycle-consistency loss, etc.
Pretty excited for this project!
This morning I got another paper (on DualGAN) and my final project: coding a dual learning GAN to translate thermal images to RGB as well as comparing how the results differ when using unpaired/unlabeled inputs versus paired/labeled inputs. Paired/labeled data will typically result in higher accuracy, however in reality, obtaining large amounts of it can be time-consuming, difficult, or impossible. So unpaired/unlabeled data is preferable in many applications as long as the accuracy isn’t too much lower. Besides the usual adversarial loss, a dual learning GAN gets around the problem of unpaired/unlabeled inputs by using a cycle-consistency loss or, as the DualGAN paper called it, a reconstruction loss. You can think of it as translating something forward and back and comparing how close the result is to the original. So there are two pairs of generators and discriminators at work rather than just one. I think that after I compare the unpaired and paired input settings, I’ll be comparing my code to another one.
I’m now in the reading room as the computers in there have a greater capacity than the ones functioning in the lab. I'm working on the model with the unpaired inputs first, and then I'll alter it for the paired inputs. Right now I’m coding the generators. They have a U-Net structure, which means that they have skip connections in the layers that allow low-level information to move directly across the network instead of all filtering through the bottleneck in the middle. It’s really interesting to see how this project is building off the papers I read during the first week, since they had information on U-Nets, cycle-consistency loss, etc.
 |
| U-Net Structure |
Pretty excited for this project!
Comments
Post a Comment