![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1142837%2Fc94574d2b43943fcc00375571b33da25%2FScreenshot%202019-07-02%20at%206.52.00%20AM.png?generation=1562030595295052&alt=media\")
These are the two set of images generated by two different GANs.\n1st One : \n
2nd One :\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1142837%2F9f254b8bcf2b2eed2f7b775aa8f58517%2FScreenshot%202019-07-02%20at%206.52.09%20AM.png?generation=1562030590544816&alt=media\")
To me first one looks more realistic dog images than second one, still second has better competition rank.
\n\nDoes that means metrics care more about similar distribution than realism ?
\n\nPhoto Credit - Dogs show off thread
", "messages": [ { "id": "566265", "postDate": "07/02/2019 01:25:40", "content": "These are the two set of images generated by two different GANs.\n1st One : \n
2nd One :\n
To me first one looks more realistic dog images than second one, still second has better competition rank.
\n\nDoes that means metrics care more about similar distribution than realism ?
\n\nPhoto Credit - Dogs show off thread
", "rawMarkdown": "These are the two set of images generated by two different GANs.\n1st One : \n\n\n2nd One :\n\n\nTo me first one looks more realistic dog images than second one, still second has better competition rank.\n\nDoes that means metrics care more about similar distribution than realism ?\n\nPhoto Credit - Dogs show off thread", "votes": null }, { "id": "566280", "postDate": "07/02/2019 01:59:31", "content": "Interesting, first one was more appealing to eyes than the second one clearly but maybe the second one is more appealing to scoreboard.
", "rawMarkdown": "Interesting, first one was more appealing to eyes than the second one clearly but maybe the second one is more appealing to scoreboard.", "votes": null }, { "id": "566335", "postDate": "07/02/2019 04:25:59", "content": "exactly !
", "rawMarkdown": "exactly !", "votes": null }, { "id": "566364", "postDate": "07/02/2019 05:07:16", "content": "Thanks for sharing. I did quite a bit of testing (so did the authors of FID), but still find these cases from time to time. Note that we use different NN and private data for the private leaderboard too, so they may have different orders in the private LB.
", "rawMarkdown": "Thanks for sharing. I did quite a bit of testing (so did the authors of FID), but still find these cases from time to time. Note that we use different NN and private data for the private leaderboard too, so they may have different orders in the private LB.", "votes": null }, { "id": "566581", "postDate": "07/02/2019 10:46:19", "content": "May I ask where you got the scores for those two sets?\nI find it rather hard to believe that the second really has a better score.
", "rawMarkdown": "May I ask where you got the scores for those two sets?\nI find it rather hard to believe that the second really has a better score.", "votes": null }, { "id": "566714", "postDate": "07/02/2019 13:58:04", "content": "Could you please post the score for each submission? Is it a big difference?
", "rawMarkdown": "Could you please post the score for each submission? Is it a big difference?", "votes": null }, { "id": "566847", "postDate": "07/02/2019 16:47:33", "content": "Maybe the first GAN is loverfitting, i mean is learning to create puppies just by replicating the input ones but its not creating new puppies images which is the objective of the GANS
", "rawMarkdown": "Maybe the first GAN is loverfitting, i mean is learning to create puppies just by replicating the input ones but its not creating new puppies images which is the objective of the GANS", "votes": null } ], "comments": [ { "id": 566280, "author_name": "gunishj", "author_url": "", "post_date": "07/02/2019 01:59:31", "content": "Interesting, first one was more appealing to eyes than the second one clearly but maybe the second one is more appealing to scoreboard.
", "votes": null, "replies": [ { "id": 566335, "author_name": "whizzkid", "author_url": "", "post_date": "07/02/2019 04:25:59", "content": "exactly !
", "votes": null, "replies": [] } ] }, { "id": 566364, "author_name": "wendykan", "author_url": "", "post_date": "07/02/2019 05:07:16", "content": "Thanks for sharing. I did quite a bit of testing (so did the authors of FID), but still find these cases from time to time. Note that we use different NN and private data for the private leaderboard too, so they may have different orders in the private LB.
", "votes": null, "replies": [] }, { "id": 566581, "author_name": "ilu000", "author_url": "", "post_date": "07/02/2019 10:46:19", "content": "May I ask where you got the scores for those two sets?\nI find it rather hard to believe that the second really has a better score.
", "votes": null, "replies": [] }, { "id": 566714, "author_name": "felipefonte99", "author_url": "", "post_date": "07/02/2019 13:58:04", "content": "Could you please post the score for each submission? Is it a big difference?
", "votes": null, "replies": [] }, { "id": 566847, "author_name": "enric1296", "author_url": "", "post_date": "07/02/2019 16:47:33", "content": "Maybe the first GAN is loverfitting, i mean is learning to create puppies just by replicating the input ones but its not creating new puppies images which is the objective of the GANS
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "566265": "These are the two set of images generated by two different GANs.\n1st One : \n\n\n2nd One :\n\n\nTo me first one looks more realistic dog images than second one, still second has better competition rank.\n\nDoes that means metrics care more about similar distribution than realism ?\n\nPhoto Credit - Dogs show off thread", "566280": "Interesting, first one was more appealing to eyes than the second one clearly but maybe the second one is more appealing to scoreboard.", "566335": "exactly !", "566364": "Thanks for sharing. I did quite a bit of testing (so did the authors of FID), but still find these cases from time to time. Note that we use different NN and private data for the private leaderboard too, so they may have different orders in the private LB.", "566581": "May I ask where you got the scores for those two sets?\nI find it rather hard to believe that the second really has a better score.", "566714": "Could you please post the score for each submission? Is it a big difference?", "566847": "Maybe the first GAN is loverfitting, i mean is learning to create puppies just by replicating the input ones but its not creating new puppies images which is the objective of the GANS" }, "source": "meta" }