![\"\"](\"https://raw.githubusercontent.com/dimitreOliveira/MachineLearning/master/Kaggle/I%E2%80%99m%20Something%20of%20a%20Painter%20Myself/photo_cycle.png\")
One thing that I think is very important when using CycleGAN is to evaluate/inspect all generator models, including their cycle performance, but yet I have not seen people doing it in public kernels, here is an example from my public kernel Improving CycleGAN - Monet paintings:
\n
![\"\"](\"https://raw.githubusercontent.com/dimitreOliveira/MachineLearning/master/Kaggle/I%E2%80%99m%20Something%20of%20a%20Painter%20Myself/monet_cycle.png\")
Here we can see that the cycled image is very close to the original image, maybe in this case the cycle consistency constraint is too strong and is making the first generated image not so good?
", "messages": [ { "id": 1057260, "postDate": "2020-10-22T14:18:32.067Z", "content": "One thing that I think is very important when using CycleGAN is to evaluate/inspect all generator models, including their cycle performance, but yet I have not seen people doing it in public kernels, here is an example from my public kernel Improving CycleGAN - Monet paintings:
\n
Here we can see that the cycled image is very close to the original image, maybe in this case the cycle consistency constraint is too strong and is making the first generated image not so good?
", "rawMarkdown": "One thing that I think is very important when using CycleGAN is to evaluate/inspect all generator models, including their cycle performance, but yet I have not seen people doing it in public kernels, here is an example from my public kernel [Improving CycleGAN - Monet paintings](https://www.kaggle.com/dimitreoliveira/improving-cyclegan-monet-paintings#Evaluating-generator-models):\n\n### Photo (input) -> Monet (generated) -> Photo (generated)\n\n\n\n### Monet (input) -> Photo (generated) -> Monet (generated)\n\n\n\nHere we can see that the cycled image is very close to the original image, maybe in this case the cycle consistency constraint is too strong and is making the first generated image not so good?", "votes": 1 } ], "comments": [], "raw_markdown_by_id": { "1057260": "One thing that I think is very important when using CycleGAN is to evaluate/inspect all generator models, including their cycle performance, but yet I have not seen people doing it in public kernels, here is an example from my public kernel [Improving CycleGAN - Monet paintings](https://www.kaggle.com/dimitreoliveira/improving-cyclegan-monet-paintings#Evaluating-generator-models):\n\n### Photo (input) -> Monet (generated) -> Photo (generated)\n\n\n\n### Monet (input) -> Photo (generated) -> Monet (generated)\n\n\n\nHere we can see that the cycled image is very close to the original image, maybe in this case the cycle consistency constraint is too strong and is making the first generated image not so good?" } }