![\"Public](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F473234%2F0d0f11a4616f72d360df1076ad1c4932%2F2020-07-07_PublicLB.png?generation=1594091294339914&alt=media\")
After long long stuck, now @daizutabi have achieved 0.57 on LB, Congrats !
\n\n
\nI would be happy if my data or baseline contributes to this break even just a little :)
After long long stuck, now @daizutabi have achieved 0.57 on LB, Congrats !
\n\n
\nI would be happy if my data or baseline contributes to this break even just a little :)
Do you guys feel like there is something wrong with this competition?
\n\n~140 people grouped up with the same score 0.56 at the top basically.
", "rawMarkdown": "Do you guys feel like there is something wrong with this competition? \n\n~140 people grouped up with the same score 0.56 at the top basically.", "votes": null }, { "id": "918180", "postDate": "07/07/2020 04:09:59", "content": "First of all, thank you for sharing your data and baseline.\nYes, I read your inference notebook and just use the threshold of 0.8 in my model.
\n\nThank you for information !
", "rawMarkdown": "Thank you for information !", "votes": null }, { "id": "918739", "postDate": "07/07/2020 13:04:30", "content": "Not really. @hidehisaarai1213 significantly contributed by sharing his work, which achieves 0.56, it's thus easily reproducible.
\n\nHowever, given that the sample submission achieves 0.54, I have a feeling it will get much higher than 0.56.
", "rawMarkdown": "Not really. @hidehisaarai1213 significantly contributed by sharing his work, which achieves 0.56, it's thus easily reproducible.\n\nHowever, given that the sample submission achieves 0.54, I have a feeling it will get much higher than 0.56.", "votes": null }, { "id": "919079", "postDate": "07/07/2020 17:39:49", "content": "Wow!
", "rawMarkdown": "Wow!", "votes": null }, { "id": "920753", "postDate": "07/08/2020 19:22:05", "content": "I thought the same thing, and am glad to not be with the crowd (as far as technique). Yet to put a true prediction together, but getting close.
", "rawMarkdown": "I thought the same thing, and am glad to not be with the crowd (as far as technique). Yet to put a true prediction together, but getting close.", "votes": null }, { "id": "920921", "postDate": "07/08/2020 23:09:09", "content": "I think this is just the beginning, there's still a lot of things left to do.
\n\nI believe we can get higher and higher 🙂
", "rawMarkdown": "I think this is just the beginning, there's still a lot of things left to do.\n\nI believe we can get higher and higher 🙂", "votes": null }, { "id": "929591", "postDate": "07/14/2020 19:10:17", "content": "Congrats!!!!!!!
", "rawMarkdown": "Congrats!!!!!!!", "votes": null }, { "id": "948210", "postDate": "07/27/2020 18:23:26", "content": "@Daizu and @ttahara thank you very much for your notebook and expanations. I myself am new here and usually use keras/tensorflow. I tried efficient-net pipeline but something does not quite stick, so I was using yours to get some sense of this a-bit-unusual dataset. I have two questions:
\n\n1) When running https://www.kaggle.com/ttahara/inference-birdsong-baseline-resnest50-fast\nwith various thresholds, I keep getting the same score 0.544 so I could not reprodu @daizutabi LB results he mentioned below.
\n\n2)More general on resnet on spectograms that I have is spectograms are essentially one-dimension per pixel so why do we use 3, wouldn't it be better to just use some grayscale version?
", "rawMarkdown": "Daizu and @ttahara thank you very much for your notebook and expanations. I myself am new here and usually use keras/tensorflow. I tried efficient-net pipeline but something does not quite stick, so I was using yours to get some sense of this a-bit-unusual dataset. I have two questions:\n\n1) When running https://www.kaggle.com/ttahara/inference-birdsong-baseline-resnest50-fast\nwith various thresholds, I keep getting the same score 0.544 so I could not reprodu @daizutabi LB results he mentioned below.\n\n2)More general on resnet on spectograms that I have is spectograms are essentially one-dimension per pixel so why do we use 3, wouldn't it be better to just use some grayscale version?", "votes": null } ], "comments": [ { "id": 918176, "author_name": "bopengiowa", "author_url": "", "post_date": "07/07/2020 03:59:15", "content": "Do you guys feel like there is something wrong with this competition?
\n\n~140 people grouped up with the same score 0.56 at the top basically.
", "votes": null, "replies": [ { "id": 918739, "author_name": "ratsimihah", "author_url": "", "post_date": "07/07/2020 13:04:30", "content": "Not really. @hidehisaarai1213 significantly contributed by sharing his work, which achieves 0.56, it's thus easily reproducible.
\n\nHowever, given that the sample submission achieves 0.54, I have a feeling it will get much higher than 0.56.
", "votes": null, "replies": [] } ] }, { "id": 918180, "author_name": "daizutabi", "author_url": "", "post_date": "07/07/2020 04:09:59", "content": "First of all, thank you for sharing your data and baseline.\nYes, I read your inference notebook and just use the threshold of 0.8 in my model.
\n\nThank you for information !
", "votes": null, "replies": [] } ] }, { "id": 919079, "author_name": "wifibobber", "author_url": "", "post_date": "07/07/2020 17:39:49", "content": "Wow!
", "votes": null, "replies": [] }, { "id": 920753, "author_name": "meckdahl", "author_url": "", "post_date": "07/08/2020 19:22:05", "content": "I thought the same thing, and am glad to not be with the crowd (as far as technique). Yet to put a true prediction together, but getting close.
", "votes": null, "replies": [ { "id": 920921, "author_name": "ttahara", "author_url": "", "post_date": "07/08/2020 23:09:09", "content": "I think this is just the beginning, there's still a lot of things left to do.
\n\nI believe we can get higher and higher 🙂
", "votes": null, "replies": [] } ] }, { "id": 929591, "author_name": "mahmudds", "author_url": "", "post_date": "07/14/2020 19:10:17", "content": "Congrats!!!!!!!
", "votes": null, "replies": [] }, { "id": 948210, "author_name": "leodav", "author_url": "", "post_date": "07/27/2020 18:23:26", "content": "@Daizu and @ttahara thank you very much for your notebook and expanations. I myself am new here and usually use keras/tensorflow. I tried efficient-net pipeline but something does not quite stick, so I was using yours to get some sense of this a-bit-unusual dataset. I have two questions:
\n\n1) When running https://www.kaggle.com/ttahara/inference-birdsong-baseline-resnest50-fast\nwith various thresholds, I keep getting the same score 0.544 so I could not reprodu @daizutabi LB results he mentioned below.
\n\n2)More general on resnet on spectograms that I have is spectograms are essentially one-dimension per pixel so why do we use 3, wouldn't it be better to just use some grayscale version?
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "918144": "After long long stuck, now @daizutabi have achieved 0.57 on LB, Congrats !\n\n\n\n