This semi-supervised challenge has been officially ended. Thank you for participating in the challenge! Especially, thank you for all your feedback in the beginning and suggestions.
\n\nIn order to learn as much as we can from your successes, I'd like to invite all of you to document what methods you tried here. By establishing common things that worked (and didn't) we can contribute the results of this competition to the community.
\n\nEvery team is welcomed to fill in the google form! We would like to hear your feedback and suggestions!
\n\nWe will be using the descriptions of the methods to select competition winners, and plan to announce both \"high-scoring\" and \"novel contribution\" awards at the FGVC Workshop at CVPR 2020. There may also be an opportunity for the selected competition winners to present their solutions virtually at the workshop.
", "messages": [ { "id": "862968", "postDate": "05/27/2020 01:53:00", "content": "This semi-supervised challenge has been officially ended. Thank you for participating in the challenge! Especially, thank you for all your feedback in the beginning and suggestions.
\n\nIn order to learn as much as we can from your successes, I'd like to invite all of you to document what methods you tried here. By establishing common things that worked (and didn't) we can contribute the results of this competition to the community.
\n\nEvery team is welcomed to fill in the google form! We would like to hear your feedback and suggestions!
\n\nWe will be using the descriptions of the methods to select competition winners, and plan to announce both \"high-scoring\" and \"novel contribution\" awards at the FGVC Workshop at CVPR 2020. There may also be an opportunity for the selected competition winners to present their solutions virtually at the workshop.
", "rawMarkdown": "This semi-supervised challenge has been officially ended. Thank you for participating in the challenge! Especially, thank you for all your feedback in the beginning and suggestions. \n\nIn order to learn as much as we can from your successes, I'd like to invite all of you to document what methods you tried [here](https://forms.gle/rWtTY91XZk97p8Ms6). By establishing common things that worked (and didn't) we can contribute the results of this competition to the community.\n\n**Every team is welcomed to fill in the google form! We would like to hear your feedback and suggestions!**\n\nWe will be using the descriptions of the methods to select competition winners, and plan to announce both \"high-scoring\" and \"novel contribution\" awards at the [FGVC Workshop](https://sites.google.com/view/fgvc7) at CVPR 2020. There may also be an opportunity for the selected competition winners to present their solutions virtually at the workshop.", "votes": null }, { "id": "863641", "postDate": "05/27/2020 13:10:58", "content": "thanks for the competition.
\n\nHopefully top teams share some hints about the hardware and time needed to get their results. In my opinion this was the real bottleneck.
\n\nSo my suggestion for the next year is if you give a computationally demanding task set a very high baseline and award the first N number of teams with some free compute.
\n\nA reasonable and strong baseline here would be max 0.30 error rate or perhaps <0.25.
\n\nMy cheap supervised approach reached 0.25 error rate (1 B7 + 1 srx50 each 0.3) and i assume 0.20 is doable with supervised only.
", "rawMarkdown": "thanks for the competition.\n\nHopefully top teams share some hints about the hardware and time needed to get their results. In my opinion this was the real bottleneck.\n\nSo my suggestion for the next year is if you give a computationally demanding task set a very high baseline and award the first N number of teams with some free compute. \n\nA reasonable and strong baseline here would be max 0.30 error rate or perhaps <0.25. \n\nMy cheap supervised approach reached 0.25 error rate (1 B7 + 1 srx50 each 0.3) and i assume 0.20 is doable with supervised only.", "votes": null }, { "id": "863971", "postDate": "05/27/2020 17:16:44", "content": "Thanks for your reply! Would you mind filling the google form so that we can know more about the details (e.g. what does B7 mean)? Thank you!!
", "rawMarkdown": "Thanks for your reply! Would you mind filling the google form so that we can know more about the details (e.g. what does B7 mean)? Thank you!!", "votes": null }, { "id": "864000", "postDate": "05/27/2020 17:35:36", "content": "But I am No 11.
", "rawMarkdown": "But I am No 11.", "votes": null }, { "id": "864007", "postDate": "05/27/2020 17:42:08", "content": "We would like every team to give us feedback! I will change the phrasing.
", "rawMarkdown": "We would like every team to give us feedback! I will change the phrasing.", "votes": null } ], "comments": [ { "id": 863641, "author_name": "valanm", "author_url": "", "post_date": "05/27/2020 13:10:58", "content": "thanks for the competition.
\n\nHopefully top teams share some hints about the hardware and time needed to get their results. In my opinion this was the real bottleneck.
\n\nSo my suggestion for the next year is if you give a computationally demanding task set a very high baseline and award the first N number of teams with some free compute.
\n\nA reasonable and strong baseline here would be max 0.30 error rate or perhaps <0.25.
\n\nMy cheap supervised approach reached 0.25 error rate (1 B7 + 1 srx50 each 0.3) and i assume 0.20 is doable with supervised only.
", "votes": null, "replies": [ { "id": 863971, "author_name": "jcfredsu", "author_url": "", "post_date": "05/27/2020 17:16:44", "content": "Thanks for your reply! Would you mind filling the google form so that we can know more about the details (e.g. what does B7 mean)? Thank you!!
", "votes": null, "replies": [] }, { "id": 864000, "author_name": "valanm", "author_url": "", "post_date": "05/27/2020 17:35:36", "content": "But I am No 11.
", "votes": null, "replies": [] }, { "id": 864007, "author_name": "jcfredsu", "author_url": "", "post_date": "05/27/2020 17:42:08", "content": "We would like every team to give us feedback! I will change the phrasing.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "862968": "This semi-supervised challenge has been officially ended. Thank you for participating in the challenge! Especially, thank you for all your feedback in the beginning and suggestions. \n\nIn order to learn as much as we can from your successes, I'd like to invite all of you to document what methods you tried [here](https://forms.gle/rWtTY91XZk97p8Ms6). By establishing common things that worked (and didn't) we can contribute the results of this competition to the community.\n\n**Every team is welcomed to fill in the google form! We would like to hear your feedback and suggestions!**\n\nWe will be using the descriptions of the methods to select competition winners, and plan to announce both \"high-scoring\" and \"novel contribution\" awards at the [FGVC Workshop](https://sites.google.com/view/fgvc7) at CVPR 2020. There may also be an opportunity for the selected competition winners to present their solutions virtually at the workshop.", "863641": "thanks for the competition.\n\nHopefully top teams share some hints about the hardware and time needed to get their results. In my opinion this was the real bottleneck.\n\nSo my suggestion for the next year is if you give a computationally demanding task set a very high baseline and award the first N number of teams with some free compute. \n\nA reasonable and strong baseline here would be max 0.30 error rate or perhaps <0.25. \n\nMy cheap supervised approach reached 0.25 error rate (1 B7 + 1 srx50 each 0.3) and i assume 0.20 is doable with supervised only.", "863971": "Thanks for your reply! Would you mind filling the google form so that we can know more about the details (e.g. what does B7 mean)? Thank you!!", "864000": "But I am No 11.", "864007": "We would like every team to give us feedback! I will change the phrasing." }, "source": "meta" }