I am new to computer vision, so far I only took part in tabular data based competitions.\nIs this competition, in your opinion a CV beginner friendly competition ?
\n\nThank you for your time!
", "messages": [ { "id": 628420, "postDate": "2019-09-17T10:00:44.593Z", "content": "@avivlevi815 \nNo Competition is for expert only. Follow the discussions, follow some of the links in discussion groups. Play with CNN and other networks. Don't be afraid to submit your solutions. There probably are some experts in this competition but there are a lot of novices that will get to the top of the LB. My own philosophy is to choose a different type of competition every time as these competitions are the best way to learn.
", "rawMarkdown": "@avivlevi815 \nNo Competition is for expert only. Follow the discussions, follow some of the links in discussion groups. Play with CNN and other networks. Don't be afraid to submit your solutions. There probably are some experts in this competition but there are a lot of novices that will get to the top of the LB. My own philosophy is to choose a different type of competition every time as these competitions are the best way to learn.", "votes": 7, "replies": [ { "id": 628539, "postDate": "2019-09-17T13:52:03.683Z", "content": "Thank you yuval !
", "rawMarkdown": "Thank you yuval !" } ] }, { "id": 634638, "postDate": "2019-09-26T14:48:22.280Z", "content": "This competition is probably one of the most difficult among CV competitions, but if you're interested then go for it. \nYou've nothing to lose and at least many thing to learn.
\n\nBut you'll likely need some descent harward or cloud computing services if you start getting really serious about the LB ranking, as Kaggle GPU kernels are not really attractive anymore.
\n\nI had the chance to participate seriously onTGS salt as my first computer vision competition, really handy and relatively small dataset. I started it just for fun , learning new things and playing with GPU kernels ( when they were highly attractuve ^^) . But as time went by, I started getting really serious on the competition, learning and applying so many different things, spending so much time on it . And finally I finished in top 100 with silver medal.
", "rawMarkdown": "This competition is probably one of the most difficult among CV competitions, but if you're interested then go for it. \nYou've nothing to lose and at least many thing to learn. \n\nBut you'll likely need some descent harward or cloud computing services if you start getting really serious about the LB ranking, as Kaggle GPU kernels are not really attractive anymore. \n\nI had the chance to participate seriously on[TGS salt](https://www.kaggle.com/c/tgs-salt-identification-challenge) as my first computer vision competition, really handy and relatively small dataset. I started it just for fun , learning new things and playing with GPU kernels ( when they were highly attractuve ^^) . But as time went by, I started getting really serious on the competition, learning and applying so many different things, spending so much time on it . And finally I finished in top 100 with silver medal. \n\n", "votes": 6, "replies": [ { "id": 634842, "postDate": "2019-09-26T20:46:52.820Z", "content": "This is a challenging task, especially for someone who relies on kernel GPUs (as my self).
\n\nI also took part in TGS as my first competition on kaggle and learned a bunch finishing in top 2%. Similarly, I had a great initial experience with recent competition APTOS. Small dataset and generous kernel GPUs at start and it turned out neither was the case at the end :). anyway learned a lot too
", "rawMarkdown": "This is a challenging task, especially for someone who relies on kernel GPUs (as my self). \n\nI also took part in TGS as my first competition on kaggle and learned a bunch finishing in top 2%. Similarly, I had a great initial experience with recent competition APTOS. Small dataset and generous kernel GPUs at start and it turned out neither was the case at the end :). anyway learned a lot too", "votes": 2 } ] }, { "id": 628558, "postDate": "2019-09-17T14:26:28.237Z", "content": "While this competition isn't restricted to anyone by any mean, this is definetly not the one I would chose to start in computer vision, as object recognition brings many more challenges proper to the detection problem. The nature of the database also makes it a bit more time consuming than a more classic problem, but that should be solved by some good notebooks. I would not call this competition beginner friendly, but if it really interests you then don't hesitate to try and read the helps and suggestions from other people. Try to visualize what you are doing at every step, and don't let yourself down if you do not succeed at first, because it may take time in this competition.
", "rawMarkdown": "While this competition isn't restricted to anyone by any mean, this is definetly not the one I would chose to start in computer vision, as object recognition brings many more challenges proper to the detection problem. The nature of the database also makes it a bit more time consuming than a more classic problem, but that should be solved by some good notebooks. I would not call this competition beginner friendly, but if it really interests you then don't hesitate to try and read the helps and suggestions from other people. Try to visualize what you are doing at every step, and don't let yourself down if you do not succeed at first, because it may take time in this competition.", "votes": 6, "replies": [ { "id": 630178, "postDate": "2019-09-19T20:56:37.830Z", "content": "I would definitely recommend any CV beginner to learn in this order:\n1. Binary with a medium-size dataset of small images (Dog vs cat, don't have anything in mind)\n2. Multiclass with a medium-size dataset of small images (e.g. MNIST)\n3. Multiclass with a small-size dataset, but this time using transfer learning \n4. Same as above, but Multilabel (APTOS 2019).\n5. Instance segmentation (Severstal 2019).\n6. Small character object detection using CenterNet (kuzushiji).\n7. Object detection from regular images (Google Open Image V4).\n8. Object detection from 3D data (This competition).
\n\nThat`s my own opinion tho; each their own preference.
", "rawMarkdown": "I would definitely recommend any CV beginner to learn in this order:\n1. Binary with a medium-size dataset of small images (Dog vs cat, don't have anything in mind)\n2. Multiclass with a medium-size dataset of small images (e.g. MNIST)\n3. Multiclass with a small-size dataset, but this time using transfer learning \n4. Same as above, but Multilabel (APTOS 2019).\n5. Instance segmentation (Severstal 2019).\n6. Small character object detection using CenterNet (kuzushiji).\n7. Object detection from regular images (Google Open Image V4).\n8. Object detection from 3D data (This competition).\n\nThat`s my own opinion tho; each their own preference.", "votes": 9 }, { "id": 631834, "postDate": "2019-09-22T19:23:40.703Z", "content": "Thank you !!
", "rawMarkdown": "Thank you !!" }, { "id": 635122, "postDate": "2019-09-27T07:19:11.587Z", "content": "@xhlulu APTOS was simple classification (for beginners), multilabel classification (for those who did not wanna give up on classification approach) and ordered regression (for those who wanted to win), 3in1 :)
", "rawMarkdown": "@xhlulu APTOS was simple classification (for beginners), multilabel classification (for those who did not wanna give up on classification approach) and ordered regression (for those who wanted to win), 3in1 :)", "votes": 2 }, { "id": 635141, "postDate": "2019-09-27T07:40:13.263Z", "content": "@valanm true, APTOS was an absolute mind fuck, one hell of an experience :p
", "rawMarkdown": "@valanm true, APTOS was an absolute mind fuck, one hell of an experience :p" } ] }, { "id": 627958, "postDate": "2019-09-16T16:29:34.207Z", "content": "I am new to computer vision, so far I only took part in tabular data based competitions.\nIs this competition, in your opinion a CV beginner friendly competition ?
\n\nThank you for your time!
", "rawMarkdown": "I am new to computer vision, so far I only took part in tabular data based competitions.\nIs this competition, in your opinion a CV beginner friendly competition ?\n\nThank you for your time!\n", "votes": 2 }, { "id": 628782, "postDate": "2019-09-17T22:19:50.897Z", "content": "I would say this competition has a higher hardware barrier than some other competitions including some other CV competitions given the dataset is 85GB.
", "rawMarkdown": "I would say this competition has a higher hardware barrier than some other competitions including some other CV competitions given the dataset is 85GB.", "votes": 2, "replies": [ { "id": 634402, "postDate": "2019-09-26T08:39:47.720Z", "content": "So, how to proceed with such amount of data without your own hardware? Is it ok to buy instance in google Cloud (or similar) for such amount of data? Will it be very expensive or not
", "rawMarkdown": "So, how to proceed with such amount of data without your own hardware? Is it ok to buy instance in google Cloud (or similar) for such amount of data? Will it be very expensive or not" } ] }, { "id": 627977, "postDate": "2019-09-16T16:49:21.317Z", "content": "@avivlevi815 \nit's good you are participation int his competition. I should say congrats to you. An definitely you will learning lots of new things in this competition.
", "rawMarkdown": "@avivlevi815 \nit's good you are participation int his competition. I should say congrats to you. An definitely you will learning lots of new things in this competition. ", "replies": [ { "id": 628018, "postDate": "2019-09-16T17:42:38.520Z", "content": "Than you sir !
", "rawMarkdown": "Than you sir !" } ] }, { "id": 628163, "postDate": "2019-09-16T21:52:38.393Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 628540, "postDate": "2019-09-17T13:52:30.807Z", "content": "Thank you !
", "rawMarkdown": "Thank you !" } ] } ], "comments": [ { "id": 628420, "author_name": "yuval reina", "author_url": "", "post_date": "2019-09-17T10:00:44.593000", "content": "@avivlevi815 \nNo Competition is for expert only. Follow the discussions, follow some of the links in discussion groups. Play with CNN and other networks. Don't be afraid to submit your solutions. There probably are some experts in this competition but there are a lot of novices that will get to the top of the LB. My own philosophy is to choose a different type of competition every time as these competitions are the best way to learn.
", "votes": 7, "replies": [ { "id": 628539, "author_name": "Aviv Levi", "author_url": "", "post_date": "2019-09-17T13:52:03.683000", "content": "Thank you yuval !
", "votes": 0, "replies": [] } ] }, { "id": 634638, "author_name": "Serigne ", "author_url": "", "post_date": "2019-09-26T14:48:22.280000", "content": "This competition is probably one of the most difficult among CV competitions, but if you're interested then go for it. \nYou've nothing to lose and at least many thing to learn.
\n\nBut you'll likely need some descent harward or cloud computing services if you start getting really serious about the LB ranking, as Kaggle GPU kernels are not really attractive anymore.
\n\nI had the chance to participate seriously onTGS salt as my first computer vision competition, really handy and relatively small dataset. I started it just for fun , learning new things and playing with GPU kernels ( when they were highly attractuve ^^) . But as time went by, I started getting really serious on the competition, learning and applying so many different things, spending so much time on it . And finally I finished in top 100 with silver medal.
", "votes": 6, "replies": [ { "id": 634842, "author_name": "Miroslav Valan", "author_url": "", "post_date": "2019-09-26T20:46:52.820000", "content": "This is a challenging task, especially for someone who relies on kernel GPUs (as my self).
\n\nI also took part in TGS as my first competition on kaggle and learned a bunch finishing in top 2%. Similarly, I had a great initial experience with recent competition APTOS. Small dataset and generous kernel GPUs at start and it turned out neither was the case at the end :). anyway learned a lot too
", "votes": 2, "replies": [] } ] }, { "id": 628558, "author_name": "Thomas Gilles", "author_url": "", "post_date": "2019-09-17T14:26:28.237000", "content": "While this competition isn't restricted to anyone by any mean, this is definetly not the one I would chose to start in computer vision, as object recognition brings many more challenges proper to the detection problem. The nature of the database also makes it a bit more time consuming than a more classic problem, but that should be solved by some good notebooks. I would not call this competition beginner friendly, but if it really interests you then don't hesitate to try and read the helps and suggestions from other people. Try to visualize what you are doing at every step, and don't let yourself down if you do not succeed at first, because it may take time in this competition.
", "votes": 6, "replies": [ { "id": 630178, "author_name": "xhlulu", "author_url": "", "post_date": "2019-09-19T20:56:37.830000", "content": "I would definitely recommend any CV beginner to learn in this order:\n1. Binary with a medium-size dataset of small images (Dog vs cat, don't have anything in mind)\n2. Multiclass with a medium-size dataset of small images (e.g. MNIST)\n3. Multiclass with a small-size dataset, but this time using transfer learning \n4. Same as above, but Multilabel (APTOS 2019).\n5. Instance segmentation (Severstal 2019).\n6. Small character object detection using CenterNet (kuzushiji).\n7. Object detection from regular images (Google Open Image V4).\n8. Object detection from 3D data (This competition).
\n\nThat`s my own opinion tho; each their own preference.
", "votes": 9, "replies": [] }, { "id": 631834, "author_name": "Aviv Levi", "author_url": "", "post_date": "2019-09-22T19:23:40.703000", "content": "Thank you !!
", "votes": 0, "replies": [] }, { "id": 635122, "author_name": "Miroslav Valan", "author_url": "", "post_date": "2019-09-27T07:19:11.587000", "content": "@xhlulu APTOS was simple classification (for beginners), multilabel classification (for those who did not wanna give up on classification approach) and ordered regression (for those who wanted to win), 3in1 :)
", "votes": 2, "replies": [] }, { "id": 635141, "author_name": "Rishabh Agrahari", "author_url": "", "post_date": "2019-09-27T07:40:13.263000", "content": "@valanm true, APTOS was an absolute mind fuck, one hell of an experience :p
", "votes": 0, "replies": [] } ] }, { "id": 628782, "author_name": "Bo Peng", "author_url": "", "post_date": "2019-09-17T22:19:50.897000", "content": "I would say this competition has a higher hardware barrier than some other competitions including some other CV competitions given the dataset is 85GB.
", "votes": 2, "replies": [ { "id": 634402, "author_name": "Artem", "author_url": "", "post_date": "2019-09-26T08:39:47.720000", "content": "So, how to proceed with such amount of data without your own hardware? Is it ok to buy instance in google Cloud (or similar) for such amount of data? Will it be very expensive or not
", "votes": 0, "replies": [] } ] }, { "id": 627977, "author_name": "Bharat Singh", "author_url": "", "post_date": "2019-09-16T16:49:21.317000", "content": "@avivlevi815 \nit's good you are participation int his competition. I should say congrats to you. An definitely you will learning lots of new things in this competition.
", "votes": 0, "replies": [ { "id": 628018, "author_name": "Aviv Levi", "author_url": "", "post_date": "2019-09-16T17:42:38.520000", "content": "Than you sir !
", "votes": 0, "replies": [] } ] }, { "id": 628163, "author_name": "", "author_url": "", "post_date": "2019-09-16T21:52:38.393000", "content": "", "votes": 0, "replies": [ { "id": 628540, "author_name": "Aviv Levi", "author_url": "", "post_date": "2019-09-17T13:52:30.807000", "content": "Thank you !
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "628420": "@avivlevi815 \nNo Competition is for expert only. Follow the discussions, follow some of the links in discussion groups. Play with CNN and other networks. Don't be afraid to submit your solutions. There probably are some experts in this competition but there are a lot of novices that will get to the top of the LB. My own philosophy is to choose a different type of competition every time as these competitions are the best way to learn.", "634638": "This competition is probably one of the most difficult among CV competitions, but if you're interested then go for it. \nYou've nothing to lose and at least many thing to learn. \n\nBut you'll likely need some descent harward or cloud computing services if you start getting really serious about the LB ranking, as Kaggle GPU kernels are not really attractive anymore. \n\nI had the chance to participate seriously on[TGS salt](https://www.kaggle.com/c/tgs-salt-identification-challenge) as my first computer vision competition, really handy and relatively small dataset. I started it just for fun , learning new things and playing with GPU kernels ( when they were highly attractuve ^^) . But as time went by, I started getting really serious on the competition, learning and applying so many different things, spending so much time on it . And finally I finished in top 100 with silver medal. \n\n", "628558": "While this competition isn't restricted to anyone by any mean, this is definetly not the one I would chose to start in computer vision, as object recognition brings many more challenges proper to the detection problem. The nature of the database also makes it a bit more time consuming than a more classic problem, but that should be solved by some good notebooks. I would not call this competition beginner friendly, but if it really interests you then don't hesitate to try and read the helps and suggestions from other people. Try to visualize what you are doing at every step, and don't let yourself down if you do not succeed at first, because it may take time in this competition.", "627958": "I am new to computer vision, so far I only took part in tabular data based competitions.\nIs this competition, in your opinion a CV beginner friendly competition ?\n\nThank you for your time!\n", "628782": "I would say this competition has a higher hardware barrier than some other competitions including some other CV competitions given the dataset is 85GB.", "627977": "@avivlevi815 \nit's good you are participation int his competition. I should say congrats to you. An definitely you will learning lots of new things in this competition. ", "628163": "" } }