\n
![\"Monet](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3741633%2F9ba7ab82c5a9c8a5636bb8bdae03f287%2Fmonet1.jpg?generation=1601144657062026&alt=media\")
\n
![\"Monet](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3741633%2F3572baeafc0bce63ded218442e5b7048%2Fmonet2.jpg?generation=1601144859264946&alt=media\")
\nImage 1 is more like painting while image 2 is just like original image with not much changes but still model produced image 2 outperforms model which produce image 1 by large margin.
I am creating painting using fast style transfer Notebook Link. I have generated different versions of models just varying intensity of style in images.
\nIf we consider below 2 images created from two different style transfer models: (These images are not from dataset, I used a high resolution image to make comparison easy)
\n
\n
\nImage 1 is more like painting while image 2 is just like original image with not much changes but still model produced image 2 outperforms model which produce image 1 by large margin.
I want to understand how evaluation works also I assumed that the objective of competition was to generate Monet style paintings but if image 2 (just like original image) is outperforming image 1, I am in doubt if my assumption about competition is wrong?
\nThanks for your time.
", "messages": [ { "id": "1028298", "postDate": "09/26/2020 18:52:53", "content": "I am creating painting using fast style transfer Notebook Link. I have generated different versions of models just varying intensity of style in images.
\nIf we consider below 2 images created from two different style transfer models: (These images are not from dataset, I used a high resolution image to make comparison easy)
\n
\n
\nImage 1 is more like painting while image 2 is just like original image with not much changes but still model produced image 2 outperforms model which produce image 1 by large margin.
I want to understand how evaluation works also I assumed that the objective of competition was to generate Monet style paintings but if image 2 (just like original image) is outperforming image 1, I am in doubt if my assumption about competition is wrong?
\nThanks for your time.
", "rawMarkdown": "I am creating painting using fast style transfer [Notebook Link](https://www.kaggle.com/tarunbisht11/generate-painting-using-fast-style-transfer). I have generated different versions of models just varying intensity of style in images. \n\nIf we consider below 2 images created from two different style transfer models: (These images are not from dataset, I used a high resolution image to make comparison easy) \n\n\nImage 1 is more like painting while image 2 is just like original image with not much changes but still model produced image 2 outperforms model which produce image 1 by large margin. \n\nI want to understand how evaluation works also I assumed that the objective of competition was to generate Monet style paintings but if image 2 (just like original image) is outperforming image 1, I am in doubt if my assumption about competition is wrong? \n\nThanks for your time.", "votes": null }, { "id": "1029074", "postDate": "09/27/2020 12:45:22", "content": "There is a pre-trained network which is used for evaluation. It compares the statistics of your generated image with the real images to compute the score. You can read more at https://www.kaggle.com/c/gan-getting-started/overview/evaluation
", "rawMarkdown": "There is a pre-trained network which is used for evaluation. It compares the statistics of your generated image with the real images to compute the score. You can read more at https://www.kaggle.com/c/gan-getting-started/overview/evaluation", "votes": null } ], "comments": [ { "id": 1029074, "author_name": "kintumiku", "author_url": "", "post_date": "09/27/2020 12:45:22", "content": "There is a pre-trained network which is used for evaluation. It compares the statistics of your generated image with the real images to compute the score. You can read more at https://www.kaggle.com/c/gan-getting-started/overview/evaluation
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1028298": "I am creating painting using fast style transfer [Notebook Link](https://www.kaggle.com/tarunbisht11/generate-painting-using-fast-style-transfer). I have generated different versions of models just varying intensity of style in images. \n\nIf we consider below 2 images created from two different style transfer models: (These images are not from dataset, I used a high resolution image to make comparison easy) \n\n\nImage 1 is more like painting while image 2 is just like original image with not much changes but still model produced image 2 outperforms model which produce image 1 by large margin. \n\nI want to understand how evaluation works also I assumed that the objective of competition was to generate Monet style paintings but if image 2 (just like original image) is outperforming image 1, I am in doubt if my assumption about competition is wrong? \n\nThanks for your time.", "1029074": "There is a pre-trained network which is used for evaluation. It compares the statistics of your generated image with the real images to compute the score. You can read more at https://www.kaggle.com/c/gan-getting-started/overview/evaluation" }, "source": "meta" }