In the previous competition, flipping flukes to create new individuals was a key to getting higher scores. However, below is my take on why it wouldn't be a good idea for this competition.
\nFlukes are symmetric, but only one view (rear side) seems to have been used in prev competition. so they are single-sided (rear side). Hence, say a spot on the right side {.\\/ } of a fluke and a spot on left { \\/.} cant be from the same fluke.
\nFins are asymmetric, but two views of the same fin can be seen in the data, hence two-sided. But not all kinds of marks would be visible from two sides.
\nCase - 1: Flip gives new individual
\nA spot on a side ->/.| when flipped is a spot on the other side of the fin |.\\<- and as the spot would exist on the other side of the fin.
Case - 2: Flip give the same individual
\nA scar on a rear edge /{ <-, when flipped, is still a scar on the rear edge ->}\\, and will be similar to the view from the other side of the fin, so can be from the same fin.
So, TLDR; it depends on the type of marking that is used to identify the individual, and it's very likely that both types of scars, often subtle, could be noticed on the fins. Hence it's better to avoid flipping.
\nPlease comment below if you think otherwise. Happy Kaggling!
", "messages": [ { "id": 1679324, "postDate": "2022-02-07T06:49:35.427Z", "content": "In the previous competition, flipping flukes to create new individuals was a key to getting higher scores. However, below is my take on why it wouldn't be a good idea for this competition.
\nFlukes are symmetric, but only one view (rear side) seems to have been used in prev competition. so they are single-sided (rear side). Hence, say a spot on the right side {.\\/ } of a fluke and a spot on left { \\/.} cant be from the same fluke.
\nFins are asymmetric, but two views of the same fin can be seen in the data, hence two-sided. But not all kinds of marks would be visible from two sides.
\nCase - 1: Flip gives new individual
\nA spot on a side ->/.| when flipped is a spot on the other side of the fin |.\\<- and as the spot would exist on the other side of the fin.
Case - 2: Flip give the same individual
\nA scar on a rear edge /{ <-, when flipped, is still a scar on the rear edge ->}\\, and will be similar to the view from the other side of the fin, so can be from the same fin.
So, TLDR; it depends on the type of marking that is used to identify the individual, and it's very likely that both types of scars, often subtle, could be noticed on the fins. Hence it's better to avoid flipping.
\nPlease comment below if you think otherwise. Happy Kaggling!
", "rawMarkdown": "In the previous competition, flipping flukes to create new individuals was a key to getting higher scores. However, below is my take on why it wouldn't be a good idea for this competition.\n\n**Flukes** are symmetric, but only one view (rear side) seems to have been used in prev competition. so they are **single-sided** (rear side). Hence, say a spot on the right side **{.\\/ }** of a fluke and a spot on left **{ \\/.}** cant be from the same fluke.\n\n**Fins** are asymmetric, but two views of the same fin can be seen in the data, hence **two-sided**. But not all kinds of marks would be visible from two sides.\n\n**Case - 1:** Flip gives new individual\nA spot on a side ->**/.|** when flipped is a spot on the other side of the fin **|.\\\\**<- and as the spot would exist on the other side of the fin.\n\n**Case - 2**: Flip give the same individual\nA scar on a rear edge **/{** <-, when flipped, is still a scar on the rear edge ->**}\\\\**, and will be similar to the view from the other side of the fin, so can be from the same fin.\n\nSo, **TLDR**; *it depends on the type of marking that is used to identify the individual*, and it's very likely that both types of scars, often subtle, could be noticed on the fins. Hence it's better to avoid flipping. \n\nPlease comment below if you think otherwise. Happy Kaggling!", "votes": 40 }, { "id": 1691616, "postDate": "2022-02-15T14:02:03.160Z", "content": "Thanks, I didn't get why flipping would give a new individual, but now I understand! Agree with all your points above.
", "rawMarkdown": "Thanks, I didn't get why flipping would give a new individual, but now I understand! Agree with all your points above. ", "votes": 1, "replies": [ { "id": 1691648, "postDate": "2022-02-15T14:33:45.613Z", "content": "Good to hear that. But just to clarify it might NOT give you a new individual in this competition
", "rawMarkdown": "Good to hear that. But just to clarify it might NOT give you a new individual in this competition", "votes": 1 } ] }, { "id": 1689137, "postDate": "2022-02-14T04:08:51.777Z", "content": "I really hadn’t thought about this… haha.
\nI was only using flipping… but now that you’ve brought this up I’m uncertain.
\nIf I have time I will run an experiment and check!
\nThanks for discussing this with everyone!
", "rawMarkdown": "I really hadn’t thought about this… haha.\n\nI was only using flipping… but now that you’ve brought this up I’m uncertain.\n\nIf I have time I will run an experiment and check!\n\nThanks for discussing this with everyone!", "votes": 1, "replies": [ { "id": 1689178, "postDate": "2022-02-14T04:27:27.493Z", "content": "You are welcome! The next big concern should be about different views and the input processing step, cant decide on downsampling vs cropping. Few interesting papers on Animal Re-id here (especially the workshop) - https://www.kaggle.com/c/happy-whale-and-dolphin/discussion/307396
", "rawMarkdown": "You are welcome! The next big concern should be about different views and the input processing step, cant decide on downsampling vs cropping. Few interesting papers on Animal Re-id here (especially the workshop) - https://www.kaggle.com/c/happy-whale-and-dolphin/discussion/307396 " } ] }, { "id": 1679403, "postDate": "2022-02-07T08:00:02.617Z", "content": "Cool! You decided to write about this \"phenomenon\"! Thank you for sharing, @bsridatta!
", "rawMarkdown": "Cool! You decided to write about this \"phenomenon\"! Thank you for sharing, @bsridatta!", "votes": 1, "replies": [ { "id": 1679417, "postDate": "2022-02-07T08:20:34.763Z", "content": "Haha, just incase some is confused like me ;) do you agree with my conclusion or am I missing something here?
", "rawMarkdown": "Haha, just incase some is confused like me ;) do you agree with my conclusion or am I missing something here? ", "votes": 1 }, { "id": 1679497, "postDate": "2022-02-07T09:00:05.203Z", "content": "The competition is only started, I think something will be shown!
", "rawMarkdown": "The competition is only started, I think something will be shown! " }, { "id": 1679715, "postDate": "2022-02-07T11:21:41.027Z", "content": "Ya I guess you should try and see what works. I am trying to find ways to easily crop the fins, and then experiment with the flips. Good luck!
", "rawMarkdown": "Ya I guess you should try and see what works. I am trying to find ways to easily crop the fins, and then experiment with the flips. Good luck!" } ] }, { "id": 1684215, "postDate": "2022-02-10T10:43:44.093Z", "content": "Did you use OpenCV, CNN or Skimage for your analytics?
", "rawMarkdown": "Did you use OpenCV, CNN or Skimage for your analytics?", "replies": [ { "id": 1686045, "postDate": "2022-02-11T18:03:11.250Z", "content": "For image augmentations you should checkout albumentations, its a package by Kaggle legends and shares some of their secret sauces
", "rawMarkdown": "For image augmentations you should checkout [albumentations](https://albumentations.ai), its a package by Kaggle legends and shares some of their secret sauces ", "votes": 2 }, { "id": 1691656, "postDate": "2022-02-15T14:35:58.217Z", "content": "@muhammadammarjamshed This was just some thoughts not any deductions from experiments.
", "rawMarkdown": "@muhammadammarjamshed This was just some thoughts not any deductions from experiments. ", "votes": 1 }, { "id": 1691794, "postDate": "2022-02-15T16:02:42.683Z", "content": "Thankyou @init27 and @bsridatta for your valuable responses. Will definitely go.over your suggestions and feedback
", "rawMarkdown": "Thankyou @init27 and @bsridatta for your valuable responses. Will definitely go.over your suggestions and feedback", "votes": 2 } ] }, { "id": 1682316, "postDate": "2022-02-09T04:44:42.190Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 1691616, "author_name": "wafflebufflo", "author_url": "", "post_date": "2022-02-15T14:02:03.160000", "content": "Thanks, I didn't get why flipping would give a new individual, but now I understand! Agree with all your points above.
", "votes": 1, "replies": [ { "id": 1691648, "author_name": "datta", "author_url": "", "post_date": "2022-02-15T14:33:45.613000", "content": "Good to hear that. But just to clarify it might NOT give you a new individual in this competition
", "votes": 1, "replies": [] } ] }, { "id": 1689137, "author_name": "Darien Schettler", "author_url": "", "post_date": "2022-02-14T04:08:51.777000", "content": "I really hadn’t thought about this… haha.
\nI was only using flipping… but now that you’ve brought this up I’m uncertain.
\nIf I have time I will run an experiment and check!
\nThanks for discussing this with everyone!
", "votes": 1, "replies": [ { "id": 1689178, "author_name": "datta", "author_url": "", "post_date": "2022-02-14T04:27:27.493000", "content": "You are welcome! The next big concern should be about different views and the input processing step, cant decide on downsampling vs cropping. Few interesting papers on Animal Re-id here (especially the workshop) - https://www.kaggle.com/c/happy-whale-and-dolphin/discussion/307396
", "votes": 0, "replies": [] } ] }, { "id": 1679403, "author_name": "Vadim Irtlach", "author_url": "", "post_date": "2022-02-07T08:00:02.617000", "content": "Cool! You decided to write about this \"phenomenon\"! Thank you for sharing, @bsridatta!
", "votes": 1, "replies": [ { "id": 1679417, "author_name": "datta", "author_url": "", "post_date": "2022-02-07T08:20:34.763000", "content": "Haha, just incase some is confused like me ;) do you agree with my conclusion or am I missing something here?
", "votes": 1, "replies": [] }, { "id": 1679497, "author_name": "Vadim Irtlach", "author_url": "", "post_date": "2022-02-07T09:00:05.203000", "content": "The competition is only started, I think something will be shown!
", "votes": 0, "replies": [] }, { "id": 1679715, "author_name": "datta", "author_url": "", "post_date": "2022-02-07T11:21:41.027000", "content": "Ya I guess you should try and see what works. I am trying to find ways to easily crop the fins, and then experiment with the flips. Good luck!
", "votes": 0, "replies": [] } ] }, { "id": 1684215, "author_name": "Muhammad Ammar Jamshed", "author_url": "", "post_date": "2022-02-10T10:43:44.093000", "content": "Did you use OpenCV, CNN or Skimage for your analytics?
", "votes": 0, "replies": [ { "id": 1686045, "author_name": "Sanyam Bhutani", "author_url": "", "post_date": "2022-02-11T18:03:11.250000", "content": "For image augmentations you should checkout albumentations, its a package by Kaggle legends and shares some of their secret sauces
", "votes": 2, "replies": [] }, { "id": 1691656, "author_name": "datta", "author_url": "", "post_date": "2022-02-15T14:35:58.217000", "content": "@muhammadammarjamshed This was just some thoughts not any deductions from experiments.
", "votes": 1, "replies": [] }, { "id": 1691794, "author_name": "Muhammad Ammar Jamshed", "author_url": "", "post_date": "2022-02-15T16:02:42.683000", "content": "Thankyou @init27 and @bsridatta for your valuable responses. Will definitely go.over your suggestions and feedback
", "votes": 2, "replies": [] } ] }, { "id": 1682316, "author_name": "", "author_url": "", "post_date": "2022-02-09T04:44:42.190000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1679324": "In the previous competition, flipping flukes to create new individuals was a key to getting higher scores. However, below is my take on why it wouldn't be a good idea for this competition.\n\n**Flukes** are symmetric, but only one view (rear side) seems to have been used in prev competition. so they are **single-sided** (rear side). Hence, say a spot on the right side **{.\\/ }** of a fluke and a spot on left **{ \\/.}** cant be from the same fluke.\n\n**Fins** are asymmetric, but two views of the same fin can be seen in the data, hence **two-sided**. But not all kinds of marks would be visible from two sides.\n\n**Case - 1:** Flip gives new individual\nA spot on a side ->**/.|** when flipped is a spot on the other side of the fin **|.\\\\**<- and as the spot would exist on the other side of the fin.\n\n**Case - 2**: Flip give the same individual\nA scar on a rear edge **/{** <-, when flipped, is still a scar on the rear edge ->**}\\\\**, and will be similar to the view from the other side of the fin, so can be from the same fin.\n\nSo, **TLDR**; *it depends on the type of marking that is used to identify the individual*, and it's very likely that both types of scars, often subtle, could be noticed on the fins. Hence it's better to avoid flipping. \n\nPlease comment below if you think otherwise. Happy Kaggling!", "1691616": "Thanks, I didn't get why flipping would give a new individual, but now I understand! Agree with all your points above. ", "1689137": "I really hadn’t thought about this… haha.\n\nI was only using flipping… but now that you’ve brought this up I’m uncertain.\n\nIf I have time I will run an experiment and check!\n\nThanks for discussing this with everyone!", "1679403": "Cool! You decided to write about this \"phenomenon\"! Thank you for sharing, @bsridatta!", "1684215": "Did you use OpenCV, CNN or Skimage for your analytics?", "1682316": "" } }