![\"035159746.jpg\"](\"http://cmp.felk.cvut.cz/~mishkdmy/aux/035159746.jpg\")
\n![\"7956c4e99.jpg\"](\"http://cmp.felk.cvut.cz/~mishkdmy/aux/7956c4e99.jpg\")
Some classes are incorrectly labeled in train. E.g, class w_ae53e45
\n\n035159746.jpg and 7956c4e99.jpg have clearly different proportions, although similar textures. And there is no \"3rd\" image to resolve ambiguity here.
\n\n\n
or class w_dfbf100: b812a5dd2.jpg and 038ffb4ad.jpg clearly looks like different whales to me.
\n\n![\"b812a5dd2.jpg\"](\"http://cmp.felk.cvut.cz/~mishkdmy/aux/b812a5dd2.jpg\")
\n![\"038ffb4ad.jpg\"](\"http://cmp.felk.cvut.cz/~mishkdmy/aux/038ffb4ad.jpg\")
I can confirm that the first two are the same whale, and the last two are different whales. Note on the first two photos, the older (black and white image) is of the whale when he or she was young. Some of the markings and some of the shape of the trailing edge have persisted and are recognizable
", "rawMarkdown": "I can confirm that the first two *are* the same whale, and the last two are different whales. Note on the first two photos, the older (black and white image) is of the whale when he or she was young. Some of the markings and some of the shape of the trailing edge have persisted and are recognizable", "votes": 3, "replies": [ { "id": 469587, "postDate": "2019-02-11T13:49:47.967Z", "content": "Thanks! Is it perspective that first two have quite a different shape, or does it change with aging?
", "rawMarkdown": "Thanks! Is it perspective that first two have quite a different shape, or does it change with aging?", "votes": 2 }, { "id": 469591, "postDate": "2019-02-11T13:55:37.277Z", "content": "that is perspective from the photo. They do change shape from a very young calf to a subadult, but almost all images in this set are either sub-adults or adults so we would only see major shape changes in the event of an injury. This does happen, due to entanglements with fishing gear and from ship strikes (for example, this whale \"Fingers\" who was uninjured in 2006 when first seen: https://happywhale.com/individual/7529;enc=29538 and then was hit by a ship which cut off part of the tail by 2017: https://happywhale.com/individual/7529;enc=14738)
", "rawMarkdown": "that is perspective from the photo. They do change shape from a very young calf to a subadult, but almost all images in this set are either sub-adults or adults so we would only see major shape changes in the event of an injury. This does happen, due to entanglements with fishing gear and from ship strikes (for example, this whale \"Fingers\" who was uninjured in 2006 when first seen: https://happywhale.com/individual/7529;enc=29538 and then was hit by a ship which cut off part of the tail by 2017: https://happywhale.com/individual/7529;enc=14738)", "votes": 5 }, { "id": 469600, "postDate": "2019-02-11T14:11:31.407Z", "content": "Vessels are sad story :(
", "rawMarkdown": "Vessels are sad story :(", "votes": 2 }, { "id": 469897, "postDate": "2019-02-12T02:22:04.500Z", "content": "Hi Ted,
\n\nThanks for the information. I have a few questions for the experts. First, as pertains to the competition, if there are some cases of same-Id-different-whale or same-whale-different-Id, will only one combination be considered correct?
\n\nSecond, on whale biology, how stable are the ridge patterns on the fluke over time? Also, the white and occasionally orangish color patterns we see on the dorsal face of the flukes, what is their function? Is there a sex difference in these patterns? Do they have a name? Are they used for signaling between whales? Do whales have good enough eyesight to recognize individuals by these patterns or do they use audio cues instead? I’m quite curious, I suppose I will try to look these up anyway but it does no harm asking the resident expert.
\n\nThanks for making this fascinating dataset available.
", "rawMarkdown": "Hi Ted,\n\nThanks for the information. I have a few questions for the experts. First, as pertains to the competition, if there are some cases of same-Id-different-whale or same-whale-different-Id, will only one combination be considered correct? \n\nSecond, on whale biology, how stable are the ridge patterns on the fluke over time? Also, the white and occasionally orangish color patterns we see on the dorsal face of the flukes, what is their function? Is there a sex difference in these patterns? Do they have a name? Are they used for signaling between whales? Do whales have good enough eyesight to recognize individuals by these patterns or do they use audio cues instead? I’m quite curious, I suppose I will try to look these up anyway but it does no harm asking the resident expert. \n\nThanks for making this fascinating dataset available. ", "votes": 1 }, { "id": 470737, "postDate": "2019-02-13T13:34:43.930Z", "content": "Hi. With respect to competition scoring, I am the biologist host; I am uncertain about which combination would be considered correct. However since all kaggle competition entries use the same dataset, any issues like this will affect all teams evenly.
\n\nWhale biology:
\n\n\n\n\nhow stable are the ridge patterns on the fluke over time?
\n
Generally very stable, except in accumulation of scarring. It appears that whales get the most scarring when they are young, in their first ~ 2 years, so from sub-adult through adulthood except in the event of major entanglements most of the ridge pattern will persist through their lifetimes. The trailing edge is unquestionably an excellent ID feature, though I would expect that the best algorithm will allow for some flexibility, tolerating some change to be able to pick up matches between photos of before an after an injury event, such as for example, individuals who have been substantially entangled in fishing gear.
\n\n\n\n\nAlso, the white and occasionally orangish color patterns we see on the dorsal face of the flukes, what is their function?
\n
The white is natural pigmentation, or the result of scarring. They scar white on black and black on white, usually. This usually persists over a lifetime.
\n\nThe orange is annual growth of diatoms when they are in very cold water (ie Antarctica), so is not good for persistent ID
\n\n\n\n\nIs there a sex difference in these patterns?
\n
No, and while general patterns are hereditary (ie Australian whales are much whiter than N Pacific whales, there does not appear to be recognizable features passed from mother to calf.
\n\n\n\n\nDo they have a name?
\n
There are many classification systems, and names for some features. My favorite is the starburst pattern:\nhttps://happywhale.com/individual/2601
\n\n\n\n\nAre they used for signaling between whales?
\n
Not that we know of. They definitely do use their tails to communicate, but that is with an acoustic show of strength, pounding their tails on the water's surface (aka \"fluke slapping\")
\n\n\n\n\nDo whales have good enough eyesight to recognize individuals by these patterns or do they use audio cues instead?
\n
Uncertain about recognition by eyesight but certainly they do recognize each other and audio is more likely.
\n\nBest of luck with your entries!\nTed
", "rawMarkdown": "Hi. With respect to competition scoring, I am the biologist host; I am uncertain about which combination would be considered correct. However since all kaggle competition entries use the same dataset, any issues like this will affect all teams evenly.\n\nWhale biology:\n>how stable are the ridge patterns on the fluke over time? \n\nGenerally very stable, except in accumulation of scarring. It appears that whales get the most scarring when they are young, in their first ~ 2 years, so from sub-adult through adulthood except in the event of major entanglements most of the ridge pattern will persist through their lifetimes. The trailing edge is unquestionably an excellent ID feature, though I would expect that the best algorithm will allow for some flexibility, tolerating some change to be able to pick up matches between photos of before an after an injury event, such as for example, individuals who have been substantially entangled in fishing gear.\n\n>Also, the white and occasionally orangish color patterns we see on the dorsal face of the flukes, what is their function? \n\nThe white is natural pigmentation, or the result of scarring. They scar white on black and black on white, usually. This usually persists over a lifetime.\n\nThe orange is annual growth of diatoms when they are in very cold water (ie Antarctica), so is *not* good for persistent ID\n\n>Is there a sex difference in these patterns? \n\nNo, and while general patterns are hereditary (ie Australian whales are much whiter than N Pacific whales, there does not appear to be recognizable features passed from mother to calf.\n\n>Do they have a name? \n\nThere are many classification systems, and names for some features. My favorite is the starburst pattern:\nhttps://happywhale.com/individual/2601\n\n> Are they used for signaling between whales? \n\nNot that we know of. They definitely do use their tails to communicate, but that is with an acoustic show of strength, pounding their tails on the water's surface (aka \"fluke slapping\")\n\n>Do whales have good enough eyesight to recognize individuals by these patterns or do they use audio cues instead? \n\nUncertain about recognition by eyesight but certainly they do recognize each other and audio is more likely.\n\nBest of luck with your entries!\nTed\n", "votes": 4 } ] }, { "id": 469535, "postDate": "2019-02-11T11:35:51.163Z", "content": "Some classes are incorrectly labeled in train. E.g, class w_ae53e45
\n\n035159746.jpg and 7956c4e99.jpg have clearly different proportions, although similar textures. And there is no \"3rd\" image to resolve ambiguity here.
\n\n\n
or class w_dfbf100: b812a5dd2.jpg and 038ffb4ad.jpg clearly looks like different whales to me.
\n\n\n
here are some samples (in red) i think label are wrong. I manually inspect all the 5004 classes. The other labels are all correct. Error label is less than 0.01% (1/10000)
\n\n![\"enter](\"https://storage.googleapis.com/kaggle-forum-message-attachments/472153/11310/Slide4.png\")
![\"enter](\"https://storage.googleapis.com/kaggle-forum-message-attachments/472153/11309/Slide3.png\")
Adding to this, here is an example of (I think, I'm no whale biologist) the same whale with different labels.
\n\nw_7537a43 has two images: 4429bede1.jpg, ada9e755a.jpg .
\n\nw_2ed0d2f has seven images (here are two): 1b8dbfabf.jpg, 422992064.jpg .
", "rawMarkdown": "Adding to this, here is an example of (I think, I'm no whale biologist) the same whale with different labels.\n\nw_7537a43 has two images: 4429bede1.jpg, ada9e755a.jpg .\n\nw_2ed0d2f has seven images (here are two): 1b8dbfabf.jpg, 422992064.jpg .\n", "replies": [ { "id": 470611, "postDate": "2019-02-13T08:47:24.047Z", "content": "Two more:
\n\nw_c97ad94 - has 8 images
\n\nw_a3181a0 - has 7 images
\n\nMy model puts the one with more images ahead of the one with fewer.
", "rawMarkdown": "Two more:\n\nw_c97ad94 - has 8 images\n\nw_a3181a0 - has 7 images\n\nMy model puts the one with more images ahead of the one with fewer. " } ] }, { "id": 470595, "postDate": "2019-02-13T08:27:52.797Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 469583, "author_name": "Ted Cheeseman", "author_url": "", "post_date": "2019-02-11T13:46:24.980000", "content": "I can confirm that the first two are the same whale, and the last two are different whales. Note on the first two photos, the older (black and white image) is of the whale when he or she was young. Some of the markings and some of the shape of the trailing edge have persisted and are recognizable
", "votes": 3, "replies": [ { "id": 469587, "author_name": "old-ufo", "author_url": "", "post_date": "2019-02-11T13:49:47.967000", "content": "Thanks! Is it perspective that first two have quite a different shape, or does it change with aging?
", "votes": 2, "replies": [] }, { "id": 469591, "author_name": "Ted Cheeseman", "author_url": "", "post_date": "2019-02-11T13:55:37.277000", "content": "that is perspective from the photo. They do change shape from a very young calf to a subadult, but almost all images in this set are either sub-adults or adults so we would only see major shape changes in the event of an injury. This does happen, due to entanglements with fishing gear and from ship strikes (for example, this whale \"Fingers\" who was uninjured in 2006 when first seen: https://happywhale.com/individual/7529;enc=29538 and then was hit by a ship which cut off part of the tail by 2017: https://happywhale.com/individual/7529;enc=14738)
", "votes": 5, "replies": [] }, { "id": 469600, "author_name": "old-ufo", "author_url": "", "post_date": "2019-02-11T14:11:31.407000", "content": "Vessels are sad story :(
", "votes": 2, "replies": [] }, { "id": 469897, "author_name": "interneuron", "author_url": "", "post_date": "2019-02-12T02:22:04.500000", "content": "Hi Ted,
\n\nThanks for the information. I have a few questions for the experts. First, as pertains to the competition, if there are some cases of same-Id-different-whale or same-whale-different-Id, will only one combination be considered correct?
\n\nSecond, on whale biology, how stable are the ridge patterns on the fluke over time? Also, the white and occasionally orangish color patterns we see on the dorsal face of the flukes, what is their function? Is there a sex difference in these patterns? Do they have a name? Are they used for signaling between whales? Do whales have good enough eyesight to recognize individuals by these patterns or do they use audio cues instead? I’m quite curious, I suppose I will try to look these up anyway but it does no harm asking the resident expert.
\n\nThanks for making this fascinating dataset available.
", "votes": 1, "replies": [] }, { "id": 470737, "author_name": "Ted Cheeseman", "author_url": "", "post_date": "2019-02-13T13:34:43.930000", "content": "Hi. With respect to competition scoring, I am the biologist host; I am uncertain about which combination would be considered correct. However since all kaggle competition entries use the same dataset, any issues like this will affect all teams evenly.
\n\nWhale biology:
\n\n\n\n\nhow stable are the ridge patterns on the fluke over time?
\n
Generally very stable, except in accumulation of scarring. It appears that whales get the most scarring when they are young, in their first ~ 2 years, so from sub-adult through adulthood except in the event of major entanglements most of the ridge pattern will persist through their lifetimes. The trailing edge is unquestionably an excellent ID feature, though I would expect that the best algorithm will allow for some flexibility, tolerating some change to be able to pick up matches between photos of before an after an injury event, such as for example, individuals who have been substantially entangled in fishing gear.
\n\n\n\n\nAlso, the white and occasionally orangish color patterns we see on the dorsal face of the flukes, what is their function?
\n
The white is natural pigmentation, or the result of scarring. They scar white on black and black on white, usually. This usually persists over a lifetime.
\n\nThe orange is annual growth of diatoms when they are in very cold water (ie Antarctica), so is not good for persistent ID
\n\n\n\n\nIs there a sex difference in these patterns?
\n
No, and while general patterns are hereditary (ie Australian whales are much whiter than N Pacific whales, there does not appear to be recognizable features passed from mother to calf.
\n\n\n\n\nDo they have a name?
\n
There are many classification systems, and names for some features. My favorite is the starburst pattern:\nhttps://happywhale.com/individual/2601
\n\n\n\n\nAre they used for signaling between whales?
\n
Not that we know of. They definitely do use their tails to communicate, but that is with an acoustic show of strength, pounding their tails on the water's surface (aka \"fluke slapping\")
\n\n\n\n\nDo whales have good enough eyesight to recognize individuals by these patterns or do they use audio cues instead?
\n
Uncertain about recognition by eyesight but certainly they do recognize each other and audio is more likely.
\n\nBest of luck with your entries!\nTed
", "votes": 4, "replies": [] } ] }, { "id": 472153, "author_name": "hengck23", "author_url": "", "post_date": "2019-02-15T12:28:30.520000", "content": "here are some samples (in red) i think label are wrong. I manually inspect all the 5004 classes. The other labels are all correct. Error label is less than 0.01% (1/10000)
\n\n
Adding to this, here is an example of (I think, I'm no whale biologist) the same whale with different labels.
\n\nw_7537a43 has two images: 4429bede1.jpg, ada9e755a.jpg .
\n\nw_2ed0d2f has seven images (here are two): 1b8dbfabf.jpg, 422992064.jpg .
", "votes": 0, "replies": [ { "id": 470611, "author_name": "interneuron", "author_url": "", "post_date": "2019-02-13T08:47:24.047000", "content": "Two more:
\n\nw_c97ad94 - has 8 images
\n\nw_a3181a0 - has 7 images
\n\nMy model puts the one with more images ahead of the one with fewer.
", "votes": 0, "replies": [] } ] }, { "id": 470595, "author_name": "", "author_url": "", "post_date": "2019-02-13T08:27:52.797000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "469583": "I can confirm that the first two *are* the same whale, and the last two are different whales. Note on the first two photos, the older (black and white image) is of the whale when he or she was young. Some of the markings and some of the shape of the trailing edge have persisted and are recognizable", "469535": "Some classes are incorrectly labeled in train. E.g, class w_ae53e45\n\n035159746.jpg and 7956c4e99.jpg have clearly different proportions, although similar textures. And there is no \"3rd\" image to resolve ambiguity here.\n\n![035159746.jpg][1]\n![7956c4e99.jpg][2]\n\nor class w_dfbf100: b812a5dd2.jpg and 038ffb4ad.jpg clearly looks like different whales to me.\n\n![b812a5dd2.jpg][3]\n![038ffb4ad.jpg][4]\n\n\n [1]: http://cmp.felk.cvut.cz/~mishkdmy/aux/035159746.jpg\n [2]: http://cmp.felk.cvut.cz/~mishkdmy/aux/7956c4e99.jpg\n [3]: http://cmp.felk.cvut.cz/~mishkdmy/aux/b812a5dd2.jpg\n [4]: http://cmp.felk.cvut.cz/~mishkdmy/aux/038ffb4ad.jpg", "472153": "here are some samples (in red) i think label are wrong. I manually inspect all the 5004 classes. The other labels are all correct. Error label is less than 0.01% (1/10000)\n\n ![enter image description here][1]\n\n\n ![enter image description here][2]\n\n\n [1]: https://storage.googleapis.com/kaggle-forum-message-attachments/472153/11310/Slide4.png\n [2]: https://storage.googleapis.com/kaggle-forum-message-attachments/472153/11309/Slide3.png", "470590": "Adding to this, here is an example of (I think, I'm no whale biologist) the same whale with different labels.\n\nw_7537a43 has two images: 4429bede1.jpg, ada9e755a.jpg .\n\nw_2ed0d2f has seven images (here are two): 1b8dbfabf.jpg, 422992064.jpg .\n", "470595": "" } }