In this competition, we're classifying animals based on wildlife camera images - with a slight twist. Both the train and test sets contain pictures of animals, but the types of animals are largely different between the two sets. They come, in fact, from completely different regions of the United States, and in different years. The challenge is to build classifiers that are robust to these differences.
\n\nExternal data is allowed for this competition - competitors are encouraged to use iNaturalist or other animal-classification data to build models. Please make sure to check out the descriptions in the Overview and Data tabs, as there are recommended models, starter code, and a number of details we don't want you to miss.
\n\nWe want to note up front that the data for this competition includes timestamps. This is required for a number of domain-specific tasks, but it can produce spurious relationships when used directly (deer may often show up at 11:23 PM in some parts of train and test, but that doesn't mean an image captured at 11:23 PM should be more likely to be a deer). Given the difficulty of generalizing between train and test sets, use care when incorporating time into your models.
\n\nAlso of note: there are some training samples with multiple animals in them. They are recorded in separate rows.
\n\nFeel free to ask questions here, and good luck!
", "messages": [ { "id": "500144", "postDate": "03/25/2019 16:43:20", "content": "In this competition, we're classifying animals based on wildlife camera images - with a slight twist. Both the train and test sets contain pictures of animals, but the types of animals are largely different between the two sets. They come, in fact, from completely different regions of the United States, and in different years. The challenge is to build classifiers that are robust to these differences.
\n\nExternal data is allowed for this competition - competitors are encouraged to use iNaturalist or other animal-classification data to build models. Please make sure to check out the descriptions in the Overview and Data tabs, as there are recommended models, starter code, and a number of details we don't want you to miss.
\n\nWe want to note up front that the data for this competition includes timestamps. This is required for a number of domain-specific tasks, but it can produce spurious relationships when used directly (deer may often show up at 11:23 PM in some parts of train and test, but that doesn't mean an image captured at 11:23 PM should be more likely to be a deer). Given the difficulty of generalizing between train and test sets, use care when incorporating time into your models.
\n\nAlso of note: there are some training samples with multiple animals in them. They are recorded in separate rows.
\n\nFeel free to ask questions here, and good luck!
", "rawMarkdown": "In this competition, we're classifying animals based on wildlife camera images - with a slight twist. Both the train and test sets contain pictures of animals, but the types of animals are largely different between the two sets. They come, in fact, from completely different regions of the United States, and in different years. The challenge is to build classifiers that are robust to these differences.\n\nExternal data is allowed for this competition - competitors are encouraged to use iNaturalist or other animal-classification data to build models. Please make sure to check out the descriptions in the Overview and Data tabs, as there are recommended models, starter code, and a number of details we don't want you to miss.\n\nWe want to note up front that the data for this competition includes timestamps. This is required for a number of domain-specific tasks, but it can produce spurious relationships when used directly (deer may often show up at 11:23 PM in some parts of train and test, but that doesn't mean an image captured at 11:23 PM should be more likely to be a deer). Given the difficulty of generalizing between train and test sets, use care when incorporating time into your models.\n\nAlso of note: there are some training samples with multiple animals in them. They are recorded in separate rows.\n\nFeel free to ask questions here, and good luck!", "votes": null }, { "id": "502087", "postDate": "03/28/2019 06:45:12", "content": "It's so hard!
", "rawMarkdown": "It's so hard!", "votes": null }, { "id": "503489", "postDate": "03/30/2019 04:05:54", "content": "What do you mean recorded in separate rows? Category_id only has one label per row, also if a test image contains more than one type of animal should we predict all that are detected? If so what is the format for multiple labels per image? Thanks.
", "rawMarkdown": "What do you mean recorded in separate rows? Category_id only has one label per row, also if a test image contains more than one type of animal should we predict all that are detected? If so what is the format for multiple labels per image? Thanks.", "votes": null }, { "id": "506279", "postDate": "04/03/2019 08:43:10", "content": "So challenging
", "rawMarkdown": "So challenging", "votes": null }, { "id": "507533", "postDate": "04/04/2019 20:39:19", "content": "Ikr...
", "rawMarkdown": "Ikr...", "votes": null }, { "id": "510407", "postDate": "04/09/2019 03:26:49", "content": "hihi~
", "rawMarkdown": "hihi~", "votes": null }, { "id": "537172", "postDate": "05/26/2019 10:43:27", "content": "hello guys, I have a little questions about the IDFG data set presented on the github webpage of this competition. What's that for? It's not mentioned that we can use that data set for training our models.
", "rawMarkdown": "hello guys, I have a little questions about the IDFG data set presented on the github webpage of this competition. What's that for? It's not mentioned that we can use that data set for training our models.", "votes": null }, { "id": "544044", "postDate": "06/05/2019 04:06:40", "content": "Thats the test set :)
", "rawMarkdown": "Thats the test set :)", "votes": null }, { "id": "546213", "postDate": "06/06/2019 11:09:25", "content": "hello, Miss / Mrs. sbeery. I have some questions about the provided detection results. \n(1). In iNat_Idaho_Detection_Results.p, the values of the key 'images' are numbers. What dose that mean? \n(2). Are the bounding box coordinates provided in a format of [x, y, width, height], with the origin at the upper left?
", "rawMarkdown": "hello, Miss / Mrs. sbeery. I have some questions about the provided detection results. \n(1). In iNat_Idaho_Detection_Results.p, the values of the key 'images' are numbers. What dose that mean? \n(2). Are the bounding box coordinates provided in a format of [x, y, width, height], with the origin at the upper left?", "votes": null }, { "id": "546233", "postDate": "06/06/2019 11:28:43", "content": "I think the numbers correspond to those in the field 'id' in 'image' in iWildCam_2019_iNat_Idaho.json. Am I right?
", "rawMarkdown": "I think the numbers correspond to those in the field 'id' in 'image' in iWildCam_2019_iNat_Idaho.json. Am I right?", "votes": null }, { "id": "546295", "postDate": "06/06/2019 12:48:55", "content": "Yes, that is correct. Please see the documentation of the output on the github page here: https://github.com/visipedia/iwildcam_comp
", "rawMarkdown": "Yes, that is correct. Please see the documentation of the output on the github page here: https://github.com/visipedia/iwildcam_comp", "votes": null } ], "comments": [ { "id": 502087, "author_name": "borings1004", "author_url": "", "post_date": "03/28/2019 06:45:12", "content": "It's so hard!
", "votes": null, "replies": [] }, { "id": 503489, "author_name": "interneuron", "author_url": "", "post_date": "03/30/2019 04:05:54", "content": "What do you mean recorded in separate rows? Category_id only has one label per row, also if a test image contains more than one type of animal should we predict all that are detected? If so what is the format for multiple labels per image? Thanks.
", "votes": null, "replies": [] }, { "id": 506279, "author_name": "myone2e", "author_url": "", "post_date": "04/03/2019 08:43:10", "content": "So challenging
", "votes": null, "replies": [ { "id": 507533, "author_name": "hyunwoolee", "author_url": "", "post_date": "04/04/2019 20:39:19", "content": "Ikr...
", "votes": null, "replies": [] }, { "id": 510407, "author_name": "myone2e", "author_url": "", "post_date": "04/09/2019 03:26:49", "content": "hihi~
", "votes": null, "replies": [] } ] }, { "id": 537172, "author_name": "taoxch", "author_url": "", "post_date": "05/26/2019 10:43:27", "content": "hello guys, I have a little questions about the IDFG data set presented on the github webpage of this competition. What's that for? It's not mentioned that we can use that data set for training our models.
", "votes": null, "replies": [ { "id": 544044, "author_name": "sbeery", "author_url": "", "post_date": "06/05/2019 04:06:40", "content": "Thats the test set :)
", "votes": null, "replies": [] } ] }, { "id": 546213, "author_name": "taoxch", "author_url": "", "post_date": "06/06/2019 11:09:25", "content": "hello, Miss / Mrs. sbeery. I have some questions about the provided detection results. \n(1). In iNat_Idaho_Detection_Results.p, the values of the key 'images' are numbers. What dose that mean? \n(2). Are the bounding box coordinates provided in a format of [x, y, width, height], with the origin at the upper left?
", "votes": null, "replies": [] }, { "id": 546233, "author_name": "taoxch", "author_url": "", "post_date": "06/06/2019 11:28:43", "content": "I think the numbers correspond to those in the field 'id' in 'image' in iWildCam_2019_iNat_Idaho.json. Am I right?
", "votes": null, "replies": [ { "id": 546295, "author_name": "sbeery", "author_url": "", "post_date": "06/06/2019 12:48:55", "content": "Yes, that is correct. Please see the documentation of the output on the github page here: https://github.com/visipedia/iwildcam_comp
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "500144": "In this competition, we're classifying animals based on wildlife camera images - with a slight twist. Both the train and test sets contain pictures of animals, but the types of animals are largely different between the two sets. They come, in fact, from completely different regions of the United States, and in different years. The challenge is to build classifiers that are robust to these differences.\n\nExternal data is allowed for this competition - competitors are encouraged to use iNaturalist or other animal-classification data to build models. Please make sure to check out the descriptions in the Overview and Data tabs, as there are recommended models, starter code, and a number of details we don't want you to miss.\n\nWe want to note up front that the data for this competition includes timestamps. This is required for a number of domain-specific tasks, but it can produce spurious relationships when used directly (deer may often show up at 11:23 PM in some parts of train and test, but that doesn't mean an image captured at 11:23 PM should be more likely to be a deer). Given the difficulty of generalizing between train and test sets, use care when incorporating time into your models.\n\nAlso of note: there are some training samples with multiple animals in them. They are recorded in separate rows.\n\nFeel free to ask questions here, and good luck!", "502087": "It's so hard!", "503489": "What do you mean recorded in separate rows? Category_id only has one label per row, also if a test image contains more than one type of animal should we predict all that are detected? If so what is the format for multiple labels per image? Thanks.", "506279": "So challenging", "507533": "Ikr...", "510407": "hihi~", "537172": "hello guys, I have a little questions about the IDFG data set presented on the github webpage of this competition. What's that for? It's not mentioned that we can use that data set for training our models.", "544044": "Thats the test set :)", "546213": "hello, Miss / Mrs. sbeery. I have some questions about the provided detection results. \n(1). In iNat_Idaho_Detection_Results.p, the values of the key 'images' are numbers. What dose that mean? \n(2). Are the bounding box coordinates provided in a format of [x, y, width, height], with the origin at the upper left?", "546233": "I think the numbers correspond to those in the field 'id' in 'image' in iWildCam_2019_iNat_Idaho.json. Am I right?", "546295": "Yes, that is correct. Please see the documentation of the output on the github page here: https://github.com/visipedia/iwildcam_comp" }, "source": "meta" }