![\"shakeup\"](\"https://i.imgur.com/AgIwWrx.png\")
We estimate there will be a pretty big shake. What's your idea?
", "messages": [ { "id": 1689018, "postDate": "2022-02-14T01:13:29.443Z", "content": "We estimate there will be a pretty big shake. What's your idea?
", "rawMarkdown": "We estimate there will be a pretty big shake. What's your idea?", "votes": 28 }, { "id": 1689104, "postDate": "2022-02-14T03:26:52.080Z", "content": "
Real World Unseen Data: AUC=0.04
", "rawMarkdown": "Real World Unseen Data: AUC=0.04", "votes": 1 } ] }, { "id": 1689060, "postDate": "2022-02-14T02:27:00.983Z", "content": "![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F3407946%2Fb674080680317a509436948ed942db23%2Fkaggle.jpg?generation=1569486219538676&alt=media\")
πππππ
", "rawMarkdown": "πππππ" } ] }, { "id": 1689490, "postDate": "2022-02-14T09:03:04.920Z", "content": "@anshulkhadse @tiquasar
\n![\"\"](\"https://i.postimg.cc/50DQ4kMF/merko-to-esa-dhak-dhak-horela-hai-768x441.jpg\")
\ncaption: heart is throbbing
\nps: only Indians know this feeling ;)
Rooting for you guys, I hope you all become masters soon.
", "rawMarkdown": "Rooting for you guys, I hope you all become masters soon.", "votes": 2 }, { "id": 1689970, "postDate": "2022-02-14T16:12:52.190Z", "content": "Thank you so much @nischaydnk :))
", "rawMarkdown": "Thank you so much @nischaydnk :))" } ] }, { "id": 1691512, "postDate": "2022-02-15T12:55:27.007Z", "content": "![\"\"](\"https://i.postimg.cc/jSwwt2Ln/65am8r.jpg\")
![\"\"](\"https://i.imgur.com/HWmPcti.png\")
The top teams may have found magic, but we couldn't find it and misunderstood that there will be a big shakeup
", "rawMarkdown": "The top teams may have found magic, but we couldn't find it and misunderstood that there will be a big shakeup", "votes": 1 }, { "id": 1689129, "postDate": "2022-02-14T04:02:26.747Z", "content": "Based on this post on TF Prize clarification:
\nFrom the hosts: \"The deployment rig contains a Jetson AGX Xavier GPU. Tensorflow engineers plan to tune the winning TF model to run efficiently on the embedded system before deploying and open sourcing. So for this challenge we want you to use a GPU for training and inference.\"
it could be that \"tuning\" would involve training further on more videos, different conditions, etc. so the private test, public test and training data are pretty similar. And Domain Generalization was not an expressed objective for this competition, trying to cater for all kinds of underwater detections.
\nHowever from the organisers' paper https://arxiv.org/abs/2111.14311 -
\n\"The dataset was collected over a single day in October 2021 at a reef in the Swain Reefs region of the GBR. There are variations in lighting, visibility, coral habitat, depth, distance from bottom and viewpoint. β¦ The released dataset consists of sequences of underwater images collected at five different areas on a reef in the GBR.\"
Single day seems good but five different areas might mean private test has things not previously seen, in different clips.
\nTLDR: Could go either way, hedge your bets in the 4 subs. Good luck all!
", "rawMarkdown": "Based on [this post on TF Prize clarification: ](https://www.kaggle.com/c/tensorflow-great-barrier-reef/discussion/291541#1671823)\nFrom the hosts: \"The deployment rig contains a Jetson AGX Xavier GPU. Tensorflow engineers plan to tune the winning TF model to run efficiently on the embedded system before deploying and open sourcing. So for this challenge we want you to use a GPU for training and inference.\"\n\nit could be that \"tuning\" would involve training further on more videos, different conditions, etc. so the private test, public test and training data are pretty similar. And Domain Generalization was not an expressed objective for this competition, trying to cater for all kinds of underwater detections.\n\nHowever from the organisers' paper https://arxiv.org/abs/2111.14311 -\n\"The dataset was collected over a single day in October 2021 at a reef in the Swain Reefs region of the GBR. There are variations in lighting, visibility, coral habitat, depth, distance from bottom and viewpoint. ... The released dataset consists of sequences of underwater images collected at five different areas on a reef in the GBR.\"\n\nSingle day seems good but five different areas might mean private test has things not previously seen, in different clips.\n\nTLDR: Could go either way, hedge your bets in the 4 subs. Good luck all! ", "votes": 1, "replies": [ { "id": 1689436, "postDate": "2022-02-14T08:19:38.540Z", "content": "I have not followed this competition to much, may I ask why Tensorflow engineers deploy model on Jetson AGX Xavier GPU, They already have their own Edge TPU?
", "rawMarkdown": "I have not followed this competition to much, may I ask why Tensorflow engineers deploy model on Jetson AGX Xavier GPU, They already have their own Edge TPU?" }, { "id": 1689657, "postDate": "2022-02-14T11:47:57.967Z", "content": "The post referenced was from Addison Howard so maybe would be able to explain more. It was relaying what the hosts had responded on the questions of Tensorflow models needing to be trained on GPU as well as inference on GPU. So gathered that was what they had and planned to use in future.
", "rawMarkdown": "The post referenced was from Addison Howard so maybe would be able to explain more. It was relaying what the hosts had responded on the questions of Tensorflow models needing to be trained on GPU as well as inference on GPU. So gathered that was what they had and planned to use in future. " }, { "id": 1689771, "postDate": "2022-02-14T13:48:47.550Z", "content": "@steamedsheep I am guessing it is small enough to fit with diver's equipment. I don't understand why training on TPU and inference on GPU is any problem since that is the main way to use TF OD API
", "rawMarkdown": "@steamedsheep I am guessing it is small enough to fit with diver's equipment. I don't understand why training on TPU and inference on GPU is any problem since that is the main way to use TF OD API" } ] }, { "id": 1689496, "postDate": "2022-02-14T09:07:48.777Z", "content": "\n", "rawMarkdown": "> Hope for the best, prepare for the worst.", "votes": 2 }, { "id": 1689026, "postDate": "2022-02-14T01:32:48.463Z", "content": "Hope for the best, prepare for the worst.
\n
Hehe sheep, you are starting a new business of shake up discussions : https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/297713
\nSartorius wasn't that bad finally. Likely another story for this one.
\nBasically is video 5 more similar to A) Videos 1-2-3 or to B) video 4 ?
\nStatistically A is more likely.
", "rawMarkdown": "Hehe sheep, you are starting a new business of shake up discussions : https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/297713\n\nSartorius wasn't that bad finally. Likely another story for this one.\n\nBasically is video 5 more similar to A) Videos 1-2-3 or to B) video 4 ?\n\nStatistically A is more likely.", "votes": 2, "replies": [ { "id": 1689438, "postDate": "2022-02-14T08:20:58.717Z", "content": "Sartorius is different, biology expriment and image sensoring is well controlled and metric is stable.
\nThis time even in trainset, the video looks different.
Seems like it was a shake up and not down for your team. Congratulations on winning @steamedsheep! ππ΅
", "rawMarkdown": "Seems like it was a shake **up** and not **down** for your team. Congratulations on winning @steamedsheep! ππ΅" }, { "id": 1690603, "postDate": "2022-02-15T02:55:00.603Z", "content": "Sheep is all you need !
", "rawMarkdown": "Sheep is all you need !" }, { "id": 1690575, "postDate": "2022-02-15T02:32:49.887Z", "content": "Really amazing. Congratulations
", "rawMarkdown": "Really amazing. Congratulations" }, { "id": 1690547, "postDate": "2022-02-15T02:03:03.183Z", "content": "Congrats master sheep! And thank you for your insights shared in this competition. A beginner like me learned a lot
", "rawMarkdown": "Congrats master sheep! And thank you for your insights shared in this competition. A beginner like me learned a lot", "replies": [ { "id": 1690562, "postDate": "2022-02-15T02:17:44.877Z", "content": "I didn't trust my CV. One of the best CV models scored 0.662 which would have been #250. Ended up not selecting it as it scored only 0.595 on public LB. Painful but necessary lesson
", "rawMarkdown": "I didn't trust my CV. One of the best CV models scored 0.662 which would have been #250. Ended up not selecting it as it scored only 0.595 on public LB. Painful but necessary lesson" } ] }, { "id": 1690468, "postDate": "2022-02-15T00:41:35.807Z", "content": "shake up is here)
\ncongrats @steamedsheep !
You are right, master sheep.
", "rawMarkdown": "You are right, master sheep." }, { "id": 1690458, "postDate": "2022-02-15T00:38:37.930Z", "content": "@steamedsheep You are great! Congratulations! π Can you share how you analyzed the Shake up?
", "rawMarkdown": "@steamedsheep You are great! Congratulations! π Can you share how you analyzed the Shake up?" }, { "id": 1689134, "postDate": "2022-02-14T04:05:45.460Z", "content": "The \"outrunner\" is so mighty, he must have find a way out
", "rawMarkdown": "The \"outrunner\" is so mighty, he must have find a way out ", "replies": [ { "id": 1689250, "postDate": "2022-02-14T05:31:21.610Z", "content": "Can't wait to learn what was used 0.013 higher than 2nd place. A true master.
", "rawMarkdown": "Can't wait to learn what was used 0.013 higher than 2nd place. A true master." }, { "id": 1689255, "postDate": "2022-02-14T05:42:48.740Z", "content": "He outran them.
", "rawMarkdown": "He outran them.", "votes": 5 } ] }, { "id": 1689082, "postDate": "2022-02-14T02:49:09.040Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 1689104, "author_name": "KhanhVD", "author_url": "", "post_date": "2022-02-14T03:26:52.080000", "content": "
Real World Unseen Data: AUC=0.04
", "votes": 1, "replies": [] } ] }, { "id": 1689060, "author_name": "Leon", "author_url": "", "post_date": "2022-02-14T02:27:00.983000", "content": "
πππππ
", "votes": 0, "replies": [] } ] }, { "id": 1689490, "author_name": "Sanchit Vijay", "author_url": "", "post_date": "2022-02-14T09:03:04.920000", "content": "@anshulkhadse @tiquasar
\n
\ncaption: heart is throbbing
\nps: only Indians know this feeling ;)
Rooting for you guys, I hope you all become masters soon.
", "votes": 2, "replies": [] }, { "id": 1689970, "author_name": "Sanchit Vijay", "author_url": "", "post_date": "2022-02-14T16:12:52.190000", "content": "Thank you so much @nischaydnk :))
", "votes": 0, "replies": [] } ] }, { "id": 1691512, "author_name": "Lime-Cake", "author_url": "", "post_date": "2022-02-15T12:55:27.007000", "content": "
The top teams may have found magic, but we couldn't find it and misunderstood that there will be a big shakeup
", "votes": 1, "replies": [] }, { "id": 1689129, "author_name": "something4kag", "author_url": "", "post_date": "2022-02-14T04:02:26.747000", "content": "Based on this post on TF Prize clarification:
\nFrom the hosts: \"The deployment rig contains a Jetson AGX Xavier GPU. Tensorflow engineers plan to tune the winning TF model to run efficiently on the embedded system before deploying and open sourcing. So for this challenge we want you to use a GPU for training and inference.\"
it could be that \"tuning\" would involve training further on more videos, different conditions, etc. so the private test, public test and training data are pretty similar. And Domain Generalization was not an expressed objective for this competition, trying to cater for all kinds of underwater detections.
\nHowever from the organisers' paper https://arxiv.org/abs/2111.14311 -
\n\"The dataset was collected over a single day in October 2021 at a reef in the Swain Reefs region of the GBR. There are variations in lighting, visibility, coral habitat, depth, distance from bottom and viewpoint. β¦ The released dataset consists of sequences of underwater images collected at five different areas on a reef in the GBR.\"
Single day seems good but five different areas might mean private test has things not previously seen, in different clips.
\nTLDR: Could go either way, hedge your bets in the 4 subs. Good luck all!
", "votes": 1, "replies": [ { "id": 1689436, "author_name": "sheep", "author_url": "", "post_date": "2022-02-14T08:19:38.540000", "content": "I have not followed this competition to much, may I ask why Tensorflow engineers deploy model on Jetson AGX Xavier GPU, They already have their own Edge TPU?
", "votes": 0, "replies": [] }, { "id": 1689657, "author_name": "something4kag", "author_url": "", "post_date": "2022-02-14T11:47:57.967000", "content": "The post referenced was from Addison Howard so maybe would be able to explain more. It was relaying what the hosts had responded on the questions of Tensorflow models needing to be trained on GPU as well as inference on GPU. So gathered that was what they had and planned to use in future.
", "votes": 0, "replies": [] }, { "id": 1689771, "author_name": "outwrest", "author_url": "", "post_date": "2022-02-14T13:48:47.550000", "content": "@steamedsheep I am guessing it is small enough to fit with diver's equipment. I don't understand why training on TPU and inference on GPU is any problem since that is the main way to use TF OD API
", "votes": 0, "replies": [] } ] }, { "id": 1689496, "author_name": "Vadim Irtlach", "author_url": "", "post_date": "2022-02-14T09:07:48.777000", "content": "\n", "votes": 2, "replies": [] }, { "id": 1689026, "author_name": "Alexandre Cadrin-ChΓͺnevert", "author_url": "", "post_date": "2022-02-14T01:32:48.463000", "content": "Hope for the best, prepare for the worst.
\n
Hehe sheep, you are starting a new business of shake up discussions : https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/297713
\nSartorius wasn't that bad finally. Likely another story for this one.
\nBasically is video 5 more similar to A) Videos 1-2-3 or to B) video 4 ?
\nStatistically A is more likely.
", "votes": 2, "replies": [ { "id": 1689438, "author_name": "sheep", "author_url": "", "post_date": "2022-02-14T08:20:58.717000", "content": "Sartorius is different, biology expriment and image sensoring is well controlled and metric is stable.
\nThis time even in trainset, the video looks different.
Seems like it was a shake up and not down for your team. Congratulations on winning @steamedsheep! ππ΅
", "votes": 0, "replies": [] }, { "id": 1690603, "author_name": "MFYfanyu", "author_url": "", "post_date": "2022-02-15T02:55:00.603000", "content": "Sheep is all you need !
", "votes": 0, "replies": [] }, { "id": 1690575, "author_name": "Robson", "author_url": "", "post_date": "2022-02-15T02:32:49.887000", "content": "Really amazing. Congratulations
", "votes": 0, "replies": [] }, { "id": 1690547, "author_name": "nymfree", "author_url": "", "post_date": "2022-02-15T02:03:03.183000", "content": "Congrats master sheep! And thank you for your insights shared in this competition. A beginner like me learned a lot
", "votes": 0, "replies": [ { "id": 1690562, "author_name": "nymfree", "author_url": "", "post_date": "2022-02-15T02:17:44.877000", "content": "I didn't trust my CV. One of the best CV models scored 0.662 which would have been #250. Ended up not selecting it as it scored only 0.595 on public LB. Painful but necessary lesson
", "votes": 0, "replies": [] } ] }, { "id": 1690468, "author_name": "Vitaly Sedelnik", "author_url": "", "post_date": "2022-02-15T00:41:35.807000", "content": "shake up is here)
\ncongrats @steamedsheep !
You are right, master sheep.
", "votes": 0, "replies": [] }, { "id": 1690458, "author_name": "Phoenix Ash", "author_url": "", "post_date": "2022-02-15T00:38:37.930000", "content": "@steamedsheep You are great! Congratulations! π Can you share how you analyzed the Shake up?
", "votes": 0, "replies": [] }, { "id": 1689134, "author_name": "Drzhuzhe", "author_url": "", "post_date": "2022-02-14T04:05:45.460000", "content": "The \"outrunner\" is so mighty, he must have find a way out
", "votes": 0, "replies": [ { "id": 1689250, "author_name": "Lime-Cake", "author_url": "", "post_date": "2022-02-14T05:31:21.610000", "content": "Can't wait to learn what was used 0.013 higher than 2nd place. A true master.
", "votes": 0, "replies": [] }, { "id": 1689255, "author_name": "Ivan Aerlic", "author_url": "", "post_date": "2022-02-14T05:42:48.740000", "content": "He outran them.
", "votes": 5, "replies": [] } ] }, { "id": 1689082, "author_name": "", "author_url": "", "post_date": "2022-02-14T02:49:09.040000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1689018": "We estimate there will be a pretty big shake. What's your idea?", "1689104": "", "1689060": "", "1689490": "@anshulkhadse @tiquasar \n\ncaption: heart is throbbing \nps: only Indians know this feeling ;)", "1691512": "", "1689218": "The top teams may have found magic, but we couldn't find it and misunderstood that there will be a big shakeup", "1689129": "Based on [this post on TF Prize clarification: ](https://www.kaggle.com/c/tensorflow-great-barrier-reef/discussion/291541#1671823)\nFrom the hosts: \"The deployment rig contains a Jetson AGX Xavier GPU. Tensorflow engineers plan to tune the winning TF model to run efficiently on the embedded system before deploying and open sourcing. So for this challenge we want you to use a GPU for training and inference.\"\n\nit could be that \"tuning\" would involve training further on more videos, different conditions, etc. so the private test, public test and training data are pretty similar. And Domain Generalization was not an expressed objective for this competition, trying to cater for all kinds of underwater detections.\n\nHowever from the organisers' paper https://arxiv.org/abs/2111.14311 -\n\"The dataset was collected over a single day in October 2021 at a reef in the Swain Reefs region of the GBR. There are variations in lighting, visibility, coral habitat, depth, distance from bottom and viewpoint. ... The released dataset consists of sequences of underwater images collected at five different areas on a reef in the GBR.\"\n\nSingle day seems good but five different areas might mean private test has things not previously seen, in different clips.\n\nTLDR: Could go either way, hedge your bets in the 4 subs. Good luck all! ", "1689496": "> Hope for the best, prepare for the worst.", "1689026": "Hehe sheep, you are starting a new business of shake up discussions : https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/297713\n\nSartorius wasn't that bad finally. Likely another story for this one.\n\nBasically is video 5 more similar to A) Videos 1-2-3 or to B) video 4 ?\n\nStatistically A is more likely.", "1690618": "Seems like it was a shake **up** and not **down** for your team. Congratulations on winning @steamedsheep! ππ΅", "1690603": "Sheep is all you need !", "1690575": "Really amazing. Congratulations", "1690547": "Congrats master sheep! And thank you for your insights shared in this competition. A beginner like me learned a lot", "1690468": "shake up is here)\ncongrats @steamedsheep !", "1690460": "You are right, master sheep.", "1690458": "@steamedsheep You are great! Congratulations! π Can you share how you analyzed the Shake up?", "1689134": "The \"outrunner\" is so mighty, he must have find a way out ", "1689082": "" } }