{ "id": 101874, "title": "How far can traditional Gan go.", "url": "/competitions/generative-dog-images/discussion/101874", "author_name": "", "post_date": "2019-07-29T08:55:56.706828500Z", "votes": 9, "comment_count": 31, "views": 0, "content": "

Long time not seeing any new comment in my previous topic What's your traditional GAN's performance?. Is everyone working on Memory Gan?\nFor my traditional Gan, we have been stuck around 40 for more than a week, I raise this topic just want to know how far can traditional Gan go? So, those who is still working on traditional Gan, could you please tell me what's your score?\n(Cause LB now does not have any meaning for me)\nThank you.

\n\n

Update: I noticed one said he achieved 20+ in this discussion., so there's still a long way to go.

", "messages": [ { "id": "586508", "postDate": "07/29/2019 08:55:56", "content": "

Long time not seeing any new comment in my previous topic What's your traditional GAN's performance?. Is everyone working on Memory Gan?\nFor my traditional Gan, we have been stuck around 40 for more than a week, I raise this topic just want to know how far can traditional Gan go? So, those who is still working on traditional Gan, could you please tell me what's your score?\n(Cause LB now does not have any meaning for me)\nThank you.

\n\n

Update: I noticed one said he achieved 20+ in this discussion., so there's still a long way to go.

", "rawMarkdown": "Long time not seeing any new comment in my previous topic [What's your traditional GAN's performance?](https://www.kaggle.com/c/generative-dog-images/discussion/99266#latest-580957). Is everyone working on Memory Gan?\nFor my traditional Gan, we have been stuck around 40 for more than a week, I raise this topic just want to know how far can traditional Gan go? So, those who is still working on traditional Gan, could you please tell me what's your score?\n(Cause LB now does not have any meaning for me)\nThank you.\n\nUpdate: I noticed one said he achieved 20+ in [this discussion.](https://www.kaggle.com/c/generative-dog-images/discussion/97753#latest-585472), so there's still a long way to go.", "votes": null }, { "id": "586529", "postDate": "07/29/2019 09:17:23", "content": "

I am still 57 with DCGAN .. Got some breakthrough last couple of days to break 60 range . Still far far away from 40 .

", "rawMarkdown": "I am still 57 with DCGAN .. Got some breakthrough last couple of days to break 60 range . Still far far away from 40 .", "votes": null }, { "id": "586532", "postDate": "07/29/2019 09:20:26", "content": "

Thanks for your information. By the way, I learn a lot from your public kernel, keep going pal!

", "rawMarkdown": "Thanks for your information. By the way, I learn a lot from your public kernel, keep going pal!", "votes": null }, { "id": "586553", "postDate": "07/29/2019 10:04:08", "content": "

my best score from DCGAN too, 've tried some other architectures but got lower scores :(

", "rawMarkdown": "my best score from DCGAN too, 've tried some other architectures but got lower scores :(", "votes": null }, { "id": "586565", "postDate": "07/29/2019 10:22:17", "content": "

I am stuck at 50s for 2 days now\n40 is a great score!

", "rawMarkdown": "I am stuck at 50s for 2 days now\n40 is a great score!", "votes": null }, { "id": "586596", "postDate": "07/29/2019 11:21:27", "content": "

stuck at score of around 60 :(

", "rawMarkdown": "stuck at score of around 60 :(", "votes": null }, { "id": "586597", "postDate": "07/29/2019 11:29:58", "content": "

Thanks for your information.👍

", "rawMarkdown": "Thanks for your information.👍", "votes": null }, { "id": "586598", "postDate": "07/29/2019 11:30:15", "content": "

Thanks for your info, keep going!

", "rawMarkdown": "Thanks for your info, keep going!", "votes": null }, { "id": "586610", "postDate": "07/29/2019 11:53:35", "content": "

52 is my best with a genuine gan. \nI’ll be extremely disappointed if kaggle allows someone forking a memoriser gan kernel to be awarded a medal of any colour. There are dozens of people working hard to improve their FID = 50-100 models and it would be a shame if that went unrewarded.

", "rawMarkdown": "52 is my best with a genuine gan. \nI’ll be extremely disappointed if kaggle allows someone forking a memoriser gan kernel to be awarded a medal of any colour. There are dozens of people working hard to improve their FID = 50-100 models and it would be a shame if that went unrewarded.", "votes": null }, { "id": "586725", "postDate": "07/29/2019 15:31:58", "content": "

40 is a great score I believe.

\n\n

Have been stuck at scores of 60 for more than a week now. \nDifferent approaches are ending up in similar ranges of FID.

\n\n

May I know, If you have trained and tuned an old GAN or used a new recent architecture?

", "rawMarkdown": "40 is a great score I believe.\n\nHave been stuck at scores of 60 for more than a week now. \nDifferent approaches are ending up in similar ranges of FID.\n\nMay I know, If you have trained and tuned an old GAN or used a new recent architecture?", "votes": null }, { "id": "586965", "postDate": "07/30/2019 00:25:00", "content": "

My score is also around 60.

", "rawMarkdown": "My score is also around 60.", "votes": null }, { "id": "588005", "postDate": "07/30/2019 02:49:27", "content": "

Thanks for your info.

", "rawMarkdown": "Thanks for your info.", "votes": null }, { "id": "588006", "postDate": "07/30/2019 02:49:58", "content": "

We tuned an old GAN.

", "rawMarkdown": "We tuned an old GAN.", "votes": null }, { "id": "588008", "postDate": "07/30/2019 02:52:15", "content": "

Keep going! Learning is more important than medals.(Although I have the same feeling with you

", "rawMarkdown": "Keep going! Learning is more important than medals.(Although I have the same feeling with you", "votes": null }, { "id": "588017", "postDate": "07/30/2019 03:11:07", "content": "

IMO , complete new GANs need more time to converge . They might yield great results later ,but not in 9 hrs .

", "rawMarkdown": "IMO , complete new GANs need more time to converge . They might yield great results later ,but not in 9 hrs .", "votes": null }, { "id": "588220", "postDate": "07/30/2019 09:56:57", "content": "

i got 57 at 10000s

", "rawMarkdown": "i got 57 at 10000s", "votes": null }, { "id": "588225", "postDate": "07/30/2019 10:00:23", "content": "

Thats cool !! Are you using DCGAN ?

", "rawMarkdown": "Thats cool !! Are you using DCGAN ?", "votes": null }, { "id": "588227", "postDate": "07/30/2019 10:02:06", "content": "

I recommend that people stop updating men gan，so we can see LB clearly

", "rawMarkdown": "I recommend that people stop updating men gan，so we can see LB clearly", "votes": null }, { "id": "588232", "postDate": "07/30/2019 10:06:57", "content": "

yes ，but i refer to a difference architecture ， i am working on it and would like to join the condition

", "rawMarkdown": "yes ，but i refer to a difference architecture ， i am working on it and would like to join the condition", "votes": null }, { "id": "588400", "postDate": "07/30/2019 14:28:00", "content": "

Ya Nirjhar, so far that has been my experience as well. \nThat's why Leon getting this score with old Gan is so encouraging in these final days.

", "rawMarkdown": "Ya Nirjhar, so far that has been my experience as well. \nThat's why Leon getting this score with old Gan is so encouraging in these final days.", "votes": null }, { "id": "588751", "postDate": "07/31/2019 01:42:23", "content": "

It's too late, I think :'(

", "rawMarkdown": "It's too late, I think :'(", "votes": null }, { "id": "588865", "postDate": "07/31/2019 06:58:51", "content": "

Nice work!

", "rawMarkdown": "Nice work!", "votes": null }, { "id": "589155", "postDate": "07/31/2019 14:36:34", "content": "

How better will traditional (generalize) GANs be ?\nMy miFID score has been stuck around 22+, who has closer or better ones ?

", "rawMarkdown": "How better will traditional (generalize) GANs be ?\nMy miFID score has been stuck around 22+, who has closer or better ones ?", "votes": null }, { "id": "589164", "postDate": "07/31/2019 14:47:27", "content": "

22 is amazing. Only the second person under 30 who has broken cover. Most people 40-70. It would have been great to see a real LB, but that isn’t going to happen.

", "rawMarkdown": "22 is amazing. Only the second person under 30 who has broken cover. Most people 40-70. It would have been great to see a real LB, but that isn’t going to happen.", "votes": null }, { "id": "589615", "postDate": "08/01/2019 06:57:24", "content": "

Amazing work👍 ! We have been stuck for more than 3 weeks. lol

", "rawMarkdown": "Amazing work👍 ! We have been stuck for more than 3 weeks. lol", "votes": null }, { "id": "589693", "postDate": "08/01/2019 08:51:16", "content": "

Impressive!

", "rawMarkdown": "Impressive!", "votes": null }, { "id": "590353", "postDate": "08/02/2019 05:54:10", "content": "

So I have achieved miFiD of 144 training Deep Convolution GAN with cropped images of dogs from the dataset. \nTraining my GAN took about 14000s and 200 epochs.\nWhich GAN architecture have you been using for this great score (miFiD 40)?

", "rawMarkdown": "So I have achieved miFiD of 144 training *Deep Convolution GAN* with cropped images of dogs from the dataset. \nTraining my GAN took about 14000s and 200 epochs.\nWhich GAN architecture have you been using for this great score (miFiD 40)?", "votes": null }, { "id": "590427", "postDate": "08/02/2019 07:19:19", "content": "

Cool! What is your distance value for this MIFID?

", "rawMarkdown": "Cool! What is your distance value for this MIFID?", "votes": null }, { "id": "590432", "postDate": "08/02/2019 07:30:00", "content": "

0.14+ now.\nThe lower miFID is, the lower cosine distance will also be.

", "rawMarkdown": "0.14+ now.\nThe lower miFID is, the lower cosine distance will also be.", "votes": null }, { "id": "591052", "postDate": "08/03/2019 04:08:43", "content": "

I am stuck at 48

", "rawMarkdown": "I am stuck at 48", "votes": null }, { "id": "591248", "postDate": "08/03/2019 11:50:31", "content": "

Same with you, DCGAN.

", "rawMarkdown": "Same with you, DCGAN.", "votes": null }, { "id": "591717", "postDate": "08/04/2019 06:29:28", "content": "

Because of great decisions of kaggle team (https://www.kaggle.com/c/generative-dog-images/discussion/102701), we need meaningful LB to make this competition competitive.

\n\n

Official LB is broken. Your posts will make this discussion substitute of it.

", "rawMarkdown": "Because of great decisions of kaggle team (https://www.kaggle.com/c/generative-dog-images/discussion/102701), we need meaningful LB to make this competition competitive.\n\nOfficial LB is broken. Your posts will make this discussion substitute of it.", "votes": null } ], "comments": [ { "id": 586529, "author_name": "phoenix9032", "author_url": "", "post_date": "07/29/2019 09:17:23", "content": "

I am still 57 with DCGAN .. Got some breakthrough last couple of days to break 60 range . Still far far away from 40 .

", "votes": null, "replies": [ { "id": 586532, "author_name": "leonshangguan", "author_url": "", "post_date": "07/29/2019 09:20:26", "content": "

Thanks for your information. By the way, I learn a lot from your public kernel, keep going pal!

", "votes": null, "replies": [] } ] }, { "id": 586553, "author_name": "ajinomoto132", "author_url": "", "post_date": "07/29/2019 10:04:08", "content": "

my best score from DCGAN too, 've tried some other architectures but got lower scores :(

", "votes": null, "replies": [] }, { "id": 586565, "author_name": "timetraveller98", "author_url": "", "post_date": "07/29/2019 10:22:17", "content": "

I am stuck at 50s for 2 days now\n40 is a great score!

", "votes": null, "replies": [ { "id": 586598, "author_name": "leonshangguan", "author_url": "", "post_date": "07/29/2019 11:30:15", "content": "

Thanks for your info, keep going!

", "votes": null, "replies": [] } ] }, { "id": 586596, "author_name": "yangsaewon", "author_url": "", "post_date": "07/29/2019 11:21:27", "content": "

stuck at score of around 60 :(

", "votes": null, "replies": [ { "id": 586597, "author_name": "leonshangguan", "author_url": "", "post_date": "07/29/2019 11:29:58", "content": "

Thanks for your information.👍

", "votes": null, "replies": [] } ] }, { "id": 586610, "author_name": "robga", "author_url": "", "post_date": "07/29/2019 11:53:35", "content": "

52 is my best with a genuine gan. \nI’ll be extremely disappointed if kaggle allows someone forking a memoriser gan kernel to be awarded a medal of any colour. There are dozens of people working hard to improve their FID = 50-100 models and it would be a shame if that went unrewarded.

", "votes": null, "replies": [ { "id": 588008, "author_name": "leonshangguan", "author_url": "", "post_date": "07/30/2019 02:52:15", "content": "

Keep going! Learning is more important than medals.(Although I have the same feeling with you

", "votes": null, "replies": [] } ] }, { "id": 586725, "author_name": "dhruuv111", "author_url": "", "post_date": "07/29/2019 15:31:58", "content": "

40 is a great score I believe.

\n\n

Have been stuck at scores of 60 for more than a week now. \nDifferent approaches are ending up in similar ranges of FID.

\n\n

May I know, If you have trained and tuned an old GAN or used a new recent architecture?

", "votes": null, "replies": [ { "id": 588006, "author_name": "leonshangguan", "author_url": "", "post_date": "07/30/2019 02:49:58", "content": "

We tuned an old GAN.

", "votes": null, "replies": [] }, { "id": 588017, "author_name": "phoenix9032", "author_url": "", "post_date": "07/30/2019 03:11:07", "content": "

IMO , complete new GANs need more time to converge . They might yield great results later ,but not in 9 hrs .

", "votes": null, "replies": [] }, { "id": 588400, "author_name": "dhruuv111", "author_url": "", "post_date": "07/30/2019 14:28:00", "content": "

Ya Nirjhar, so far that has been my experience as well. \nThat's why Leon getting this score with old Gan is so encouraging in these final days.

", "votes": null, "replies": [] } ] }, { "id": 586965, "author_name": "mochidan", "author_url": "", "post_date": "07/30/2019 00:25:00", "content": "

My score is also around 60.

", "votes": null, "replies": [ { "id": 588005, "author_name": "leonshangguan", "author_url": "", "post_date": "07/30/2019 02:49:27", "content": "

Thanks for your info.

", "votes": null, "replies": [] } ] }, { "id": 588220, "author_name": "peacemakertsp", "author_url": "", "post_date": "07/30/2019 09:56:57", "content": "

i got 57 at 10000s

", "votes": null, "replies": [ { "id": 588225, "author_name": "phoenix9032", "author_url": "", "post_date": "07/30/2019 10:00:23", "content": "

Thats cool !! Are you using DCGAN ?

", "votes": null, "replies": [] }, { "id": 588232, "author_name": "peacemakertsp", "author_url": "", "post_date": "07/30/2019 10:06:57", "content": "

yes ，but i refer to a difference architecture ， i am working on it and would like to join the condition

", "votes": null, "replies": [] }, { "id": 588865, "author_name": "leonshangguan", "author_url": "", "post_date": "07/31/2019 06:58:51", "content": "

Nice work!

", "votes": null, "replies": [] } ] }, { "id": 588227, "author_name": "peacemakertsp", "author_url": "", "post_date": "07/30/2019 10:02:06", "content": "

I recommend that people stop updating men gan，so we can see LB clearly

", "votes": null, "replies": [ { "id": 588751, "author_name": "quangbk", "author_url": "", "post_date": "07/31/2019 01:42:23", "content": "

It's too late, I think :'(

", "votes": null, "replies": [] } ] }, { "id": 589155, "author_name": "yukia18", "author_url": "", "post_date": "07/31/2019 14:36:34", "content": "

How better will traditional (generalize) GANs be ?\nMy miFID score has been stuck around 22+, who has closer or better ones ?

", "votes": null, "replies": [ { "id": 589164, "author_name": "robga", "author_url": "", "post_date": "07/31/2019 14:47:27", "content": "

22 is amazing. Only the second person under 30 who has broken cover. Most people 40-70. It would have been great to see a real LB, but that isn’t going to happen.

", "votes": null, "replies": [] }, { "id": 589615, "author_name": "leonshangguan", "author_url": "", "post_date": "08/01/2019 06:57:24", "content": "

Amazing work👍 ! We have been stuck for more than 3 weeks. lol

", "votes": null, "replies": [] }, { "id": 589693, "author_name": "yangddd", "author_url": "", "post_date": "08/01/2019 08:51:16", "content": "

Impressive!

", "votes": null, "replies": [] }, { "id": 590427, "author_name": "nazarov", "author_url": "", "post_date": "08/02/2019 07:19:19", "content": "

Cool! What is your distance value for this MIFID?

", "votes": null, "replies": [] }, { "id": 590432, "author_name": "yukia18", "author_url": "", "post_date": "08/02/2019 07:30:00", "content": "

0.14+ now.\nThe lower miFID is, the lower cosine distance will also be.

", "votes": null, "replies": [] } ] }, { "id": 590353, "author_name": "thanatoz", "author_url": "", "post_date": "08/02/2019 05:54:10", "content": "

So I have achieved miFiD of 144 training Deep Convolution GAN with cropped images of dogs from the dataset. \nTraining my GAN took about 14000s and 200 epochs.\nWhich GAN architecture have you been using for this great score (miFiD 40)?

", "votes": null, "replies": [ { "id": 591248, "author_name": "leonshangguan", "author_url": "", "post_date": "08/03/2019 11:50:31", "content": "

Same with you, DCGAN.

", "votes": null, "replies": [] } ] }, { "id": 591052, "author_name": "felipefonte99", "author_url": "", "post_date": "08/03/2019 04:08:43", "content": "

I am stuck at 48

", "votes": null, "replies": [] }, { "id": 591717, "author_name": "yukia18", "author_url": "", "post_date": "08/04/2019 06:29:28", "content": "

Because of great decisions of kaggle team (https://www.kaggle.com/c/generative-dog-images/discussion/102701), we need meaningful LB to make this competition competitive.

\n\n

Official LB is broken. Your posts will make this discussion substitute of it.

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "586508": "Long time not seeing any new comment in my previous topic [What's your traditional GAN's performance?](https://www.kaggle.com/c/generative-dog-images/discussion/99266#latest-580957). Is everyone working on Memory Gan?\nFor my traditional Gan, we have been stuck around 40 for more than a week, I raise this topic just want to know how far can traditional Gan go? So, those who is still working on traditional Gan, could you please tell me what's your score?\n(Cause LB now does not have any meaning for me)\nThank you.\n\nUpdate: I noticed one said he achieved 20+ in [this discussion.](https://www.kaggle.com/c/generative-dog-images/discussion/97753#latest-585472), so there's still a long way to go.", "586529": "I am still 57 with DCGAN .. Got some breakthrough last couple of days to break 60 range . Still far far away from 40 .", "586532": "Thanks for your information. By the way, I learn a lot from your public kernel, keep going pal!", "586553": "my best score from DCGAN too, 've tried some other architectures but got lower scores :(", "586565": "I am stuck at 50s for 2 days now\n40 is a great score!", "586596": "stuck at score of around 60 :(", "586597": "Thanks for your information.👍", "586598": "Thanks for your info, keep going!", "586610": "52 is my best with a genuine gan. \nI’ll be extremely disappointed if kaggle allows someone forking a memoriser gan kernel to be awarded a medal of any colour. There are dozens of people working hard to improve their FID = 50-100 models and it would be a shame if that went unrewarded.", "586725": "40 is a great score I believe.\n\nHave been stuck at scores of 60 for more than a week now. \nDifferent approaches are ending up in similar ranges of FID.\n\nMay I know, If you have trained and tuned an old GAN or used a new recent architecture?", "586965": "My score is also around 60.", "588005": "Thanks for your info.", "588006": "We tuned an old GAN.", "588008": "Keep going! Learning is more important than medals.(Although I have the same feeling with you", "588017": "IMO , complete new GANs need more time to converge . They might yield great results later ,but not in 9 hrs .", "588220": "i got 57 at 10000s", "588225": "Thats cool !! Are you using DCGAN ?", "588227": "I recommend that people stop updating men gan，so we can see LB clearly", "588232": "yes ，but i refer to a difference architecture ， i am working on it and would like to join the condition", "588400": "Ya Nirjhar, so far that has been my experience as well. \nThat's why Leon getting this score with old Gan is so encouraging in these final days.", "588751": "It's too late, I think :'(", "588865": "Nice work!", "589155": "How better will traditional (generalize) GANs be ?\nMy miFID score has been stuck around 22+, who has closer or better ones ?", "589164": "22 is amazing. Only the second person under 30 who has broken cover. Most people 40-70. It would have been great to see a real LB, but that isn’t going to happen.", "589615": "Amazing work👍 ! We have been stuck for more than 3 weeks. lol", "589693": "Impressive!", "590353": "So I have achieved miFiD of 144 training *Deep Convolution GAN* with cropped images of dogs from the dataset. \nTraining my GAN took about 14000s and 200 epochs.\nWhich GAN architecture have you been using for this great score (miFiD 40)?", "590427": "Cool! What is your distance value for this MIFID?", "590432": "0.14+ now.\nThe lower miFID is, the lower cosine distance will also be.", "591052": "I am stuck at 48", "591248": "Same with you, DCGAN.", "591717": "Because of great decisions of kaggle team (https://www.kaggle.com/c/generative-dog-images/discussion/102701), we need meaningful LB to make this competition competitive.\n\nOfficial LB is broken. Your posts will make this discussion substitute of it." }, "source": "meta" }