Congratulations to all the winners in this competition. Congratulate and thank my teammate Tom, earhian ,sheep and CHAN, It's memorable to team with you all.\nDo not trust the LB, trust your CV. This is what I learned in this competition.
\n\nOur Model Design \n1. binary classifier on ship/no ship (all images) -> resnet34, se-resnext50 based model\n2. unet models (ship images only) \n3. unet models with deepsupervison (all images) \n4. mask rcnn (ship images only)
\n\nOur final submission combines results of different models above, details will be updated and some of our code will be on https://github.com/SeuTao/Kaggle_Airbus2018_8th_code
\n\nPS: It's really lucky for us to survive the huge shakeup.
\n\nActually pseudo labeling worked in this competition (single model 853 in private)
", "messages": [ { "id": 421475, "postDate": "2018-11-15T03:46:09.780Z", "content": "Congratulations to all the winners in this competition. Congratulate and thank my teammate Tom, earhian ,sheep and CHAN, It's memorable to team with you all.\nDo not trust the LB, trust your CV. This is what I learned in this competition.
\n\nOur Model Design \n1. binary classifier on ship/no ship (all images) -> resnet34, se-resnext50 based model\n2. unet models (ship images only) \n3. unet models with deepsupervison (all images) \n4. mask rcnn (ship images only)
\n\nOur final submission combines results of different models above, details will be updated and some of our code will be on https://github.com/SeuTao/Kaggle_Airbus2018_8th_code
\n\nPS: It's really lucky for us to survive the huge shakeup.
\n\nActually pseudo labeling worked in this competition (single model 853 in private)
", "rawMarkdown": "Congratulations to all the winners in this competition. Congratulate and thank my teammate Tom, earhian ,sheep and CHAN, It's memorable to team with you all.\nDo not trust the LB, trust your CV. This is what I learned in this competition.\n\n\n\n\n**Our Model Design** \n1. binary classifier on ship/no ship (all images) -> resnet34, se-resnext50 based model\n2. unet models (ship images only) \n3. unet models with deepsupervison (all images) \n4. mask rcnn (ship images only)\n\n\nOur final submission combines results of different models above, details will be updated and some of our code will be on https://github.com/SeuTao/Kaggle_Airbus2018_8th_code\n\nPS: It's really lucky for us to survive the huge shakeup.\n\n Actually pseudo labeling worked in this competition (single model 853 in private)", "votes": 42 }, { "id": 421551, "postDate": "2018-11-15T05:44:22.647Z", "content": "Do not trust the LB, trust your CV. This is what I learned in this competition.
\n\nWhen looking at scores, we should learn not to look at the numbers. It is “what the number represents” that is important.
", "rawMarkdown": "Do not trust the LB, trust your CV. This is what I learned in this competition.\n\n\nWhen looking at scores, we should learn not to look at the numbers. It is “what the number represents” that is important.", "votes": 9 }, { "id": 421624, "postDate": "2018-11-15T07:51:02.603Z", "content": "Congrats to your team and many thanks for sharing your team solution! :-)
", "rawMarkdown": "Congrats to your team and many thanks for sharing your team solution! :-)", "votes": 1 }, { "id": 421497, "postDate": "2018-11-15T04:18:59.290Z", "content": "Congrats and many thanks to your shared solutions! I also learnt a lot from your TGS 4th place solution.
", "rawMarkdown": "Congrats and many thanks to your shared solutions! I also learnt a lot from your [TGS 4th place solution](https://github.com/SeuTao/Kaggle_TGS2018_4th_solution).", "votes": 1, "replies": [ { "id": 421498, "postDate": "2018-11-15T04:27:00.793Z", "content": "Thank u!
", "rawMarkdown": "Thank u! " } ] }, { "id": 421957, "postDate": "2018-11-15T15:55:01.093Z", "content": "Congratulations :D
\n\nMay I ask you what was your training scheme? Like 256, 384, 768? Resize or crops? Inference at 768? These kind of things
", "rawMarkdown": "Congratulations :D \n\nMay I ask you what was your training scheme? Like 256, 384, 768? Resize or crops? Inference at 768? These kind of things", "votes": 2, "replies": [ { "id": 421999, "postDate": "2018-11-15T16:55:00.663Z", "content": "I start to train model on 384 resized image with adam then fintune in 768.
", "rawMarkdown": "I start to train model on 384 resized image with adam then fintune in 768. ", "votes": 1 } ] }, { "id": 421844, "postDate": "2018-11-15T13:38:57.867Z", "content": "Congrats to your team on the good job on both this competition and TGS..
", "rawMarkdown": "Congrats to your team on the good job on both this competition and TGS..", "replies": [ { "id": 422002, "postDate": "2018-11-15T16:55:31.320Z", "content": "Thank u!
", "rawMarkdown": "Thank u!" } ] }, { "id": 421726, "postDate": "2018-11-15T10:31:10.443Z", "content": "Congrats , how many GPUs your team using ?
", "rawMarkdown": "Congrats , how many GPUs your team using ? ", "replies": [ { "id": 421994, "postDate": "2018-11-15T16:50:10.673Z", "content": "Threre are about 12 gpus in our team.
", "rawMarkdown": "Threre are about 12 gpus in our team." } ] }, { "id": 421631, "postDate": "2018-11-15T08:05:13.730Z", "content": "Congratulations! and thanks for sharing.\nI have a question. May I ask \"What label or output was used for the deepsupervision(3)\"?
", "rawMarkdown": "Congratulations! and thanks for sharing.\nI have a question. May I ask \"What label or output was used for the deepsupervision(3)\"?" }, { "id": 421520, "postDate": "2018-11-15T05:10:32.777Z", "content": "congrats! great work!
", "rawMarkdown": "congrats! great work!" }, { "id": 421517, "postDate": "2018-11-15T05:02:55.857Z", "content": "Congrats Tao, well deserved expert tier and master tier is definitely coming soon.
", "rawMarkdown": "Congrats Tao, well deserved expert tier and master tier is definitely coming soon." }, { "id": 421485, "postDate": "2018-11-15T04:03:48.467Z", "content": "Congratulations on becoming one of the most solid kaggle experts :)
", "rawMarkdown": "Congratulations on becoming one of the most solid kaggle experts :)", "replies": [ { "id": 421490, "postDate": "2018-11-15T04:08:24.873Z", "content": "haha, let's doodle!
", "rawMarkdown": "haha, let's doodle!", "votes": 2 } ] }, { "id": 421483, "postDate": "2018-11-15T03:59:47.803Z", "content": "Good job!
", "rawMarkdown": "Good job!", "replies": [ { "id": 421488, "postDate": "2018-11-15T04:07:05.167Z", "content": "Thank u again!
", "rawMarkdown": "Thank u again!" } ] }, { "id": 421845, "postDate": "2018-11-15T13:41:38.713Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 422000, "postDate": "2018-11-15T16:55:18.320Z", "content": "Thanks!
", "rawMarkdown": "Thanks!" } ] }, { "id": 421679, "postDate": "2018-11-15T09:05:00.450Z", "content": "Congrats and thanks for sharing.
", "rawMarkdown": "Congrats and thanks for sharing." } ], "comments": [ { "id": 421551, "author_name": "hengck23", "author_url": "", "post_date": "2018-11-15T05:44:22.647000", "content": "Do not trust the LB, trust your CV. This is what I learned in this competition.
\n\nWhen looking at scores, we should learn not to look at the numbers. It is “what the number represents” that is important.
", "votes": 9, "replies": [] }, { "id": 421624, "author_name": "", "author_url": "", "post_date": "2018-11-15T07:51:02.603000", "content": "Congrats to your team and many thanks for sharing your team solution! :-)
", "votes": 1, "replies": [] }, { "id": 421497, "author_name": "Mark Peng", "author_url": "", "post_date": "2018-11-15T04:18:59.290000", "content": "Congrats and many thanks to your shared solutions! I also learnt a lot from your TGS 4th place solution.
", "votes": 1, "replies": [ { "id": 421498, "author_name": "SeuTao", "author_url": "", "post_date": "2018-11-15T04:27:00.793000", "content": "Thank u!
", "votes": 0, "replies": [] } ] }, { "id": 421957, "author_name": "Eduardo Rocha de Andrade", "author_url": "", "post_date": "2018-11-15T15:55:01.093000", "content": "Congratulations :D
\n\nMay I ask you what was your training scheme? Like 256, 384, 768? Resize or crops? Inference at 768? These kind of things
", "votes": 2, "replies": [ { "id": 421999, "author_name": "SeuTao", "author_url": "", "post_date": "2018-11-15T16:55:00.663000", "content": "I start to train model on 384 resized image with adam then fintune in 768.
", "votes": 1, "replies": [] } ] }, { "id": 421844, "author_name": "Vishy", "author_url": "", "post_date": "2018-11-15T13:38:57.867000", "content": "Congrats to your team on the good job on both this competition and TGS..
", "votes": 0, "replies": [ { "id": 422002, "author_name": "SeuTao", "author_url": "", "post_date": "2018-11-15T16:55:31.320000", "content": "Thank u!
", "votes": 0, "replies": [] } ] }, { "id": 421726, "author_name": "soywu", "author_url": "", "post_date": "2018-11-15T10:31:10.443000", "content": "Congrats , how many GPUs your team using ?
", "votes": 0, "replies": [ { "id": 421994, "author_name": "SeuTao", "author_url": "", "post_date": "2018-11-15T16:50:10.673000", "content": "Threre are about 12 gpus in our team.
", "votes": 0, "replies": [] } ] }, { "id": 421631, "author_name": "Youhan Lee", "author_url": "", "post_date": "2018-11-15T08:05:13.730000", "content": "Congratulations! and thanks for sharing.\nI have a question. May I ask \"What label or output was used for the deepsupervision(3)\"?
", "votes": 0, "replies": [] }, { "id": 421520, "author_name": "YUNFEI DUAN", "author_url": "", "post_date": "2018-11-15T05:10:32.777000", "content": "congrats! great work!
", "votes": 0, "replies": [] }, { "id": 421517, "author_name": "Yiheng Wang", "author_url": "", "post_date": "2018-11-15T05:02:55.857000", "content": "Congrats Tao, well deserved expert tier and master tier is definitely coming soon.
", "votes": 0, "replies": [] }, { "id": 421485, "author_name": "Peiyuan Liao", "author_url": "", "post_date": "2018-11-15T04:03:48.467000", "content": "Congratulations on becoming one of the most solid kaggle experts :)
", "votes": 0, "replies": [ { "id": 421490, "author_name": "SeuTao", "author_url": "", "post_date": "2018-11-15T04:08:24.873000", "content": "haha, let's doodle!
", "votes": 2, "replies": [] } ] }, { "id": 421483, "author_name": "MarcYueZhao", "author_url": "", "post_date": "2018-11-15T03:59:47.803000", "content": "Good job!
", "votes": 0, "replies": [ { "id": 421488, "author_name": "SeuTao", "author_url": "", "post_date": "2018-11-15T04:07:05.167000", "content": "Thank u again!
", "votes": 0, "replies": [] } ] }, { "id": 421845, "author_name": "", "author_url": "", "post_date": "2018-11-15T13:41:38.713000", "content": "", "votes": 0, "replies": [ { "id": 422000, "author_name": "SeuTao", "author_url": "", "post_date": "2018-11-15T16:55:18.320000", "content": "Thanks!
", "votes": 0, "replies": [] } ] }, { "id": 421679, "author_name": "YaGana Sheriff-Hussaini", "author_url": "", "post_date": "2018-11-15T09:05:00.450000", "content": "Congrats and thanks for sharing.
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "421475": "Congratulations to all the winners in this competition. Congratulate and thank my teammate Tom, earhian ,sheep and CHAN, It's memorable to team with you all.\nDo not trust the LB, trust your CV. This is what I learned in this competition.\n\n\n\n\n**Our Model Design** \n1. binary classifier on ship/no ship (all images) -> resnet34, se-resnext50 based model\n2. unet models (ship images only) \n3. unet models with deepsupervison (all images) \n4. mask rcnn (ship images only)\n\n\nOur final submission combines results of different models above, details will be updated and some of our code will be on https://github.com/SeuTao/Kaggle_Airbus2018_8th_code\n\nPS: It's really lucky for us to survive the huge shakeup.\n\n Actually pseudo labeling worked in this competition (single model 853 in private)", "421551": "Do not trust the LB, trust your CV. This is what I learned in this competition.\n\n\nWhen looking at scores, we should learn not to look at the numbers. It is “what the number represents” that is important.", "421624": "Congrats to your team and many thanks for sharing your team solution! :-)", "421497": "Congrats and many thanks to your shared solutions! I also learnt a lot from your [TGS 4th place solution](https://github.com/SeuTao/Kaggle_TGS2018_4th_solution).", "421957": "Congratulations :D \n\nMay I ask you what was your training scheme? Like 256, 384, 768? Resize or crops? Inference at 768? These kind of things", "421844": "Congrats to your team on the good job on both this competition and TGS..", "421726": "Congrats , how many GPUs your team using ? ", "421631": "Congratulations! and thanks for sharing.\nI have a question. May I ask \"What label or output was used for the deepsupervision(3)\"?", "421520": "congrats! great work!", "421517": "Congrats Tao, well deserved expert tier and master tier is definitely coming soon.", "421485": "Congratulations on becoming one of the most solid kaggle experts :)", "421483": "Good job!", "421845": "", "421679": "Congrats and thanks for sharing." } }