Hi,
\n\nThe following classes in the human labels have multiple keys that map to them. Is this an issue?. I guess predicting either of the keys then is counted as correct?
\n\n['Bun', 'String instrument', 'Organ', 'Cape', 'Ball', 'Runway', 'Dome', 'Zombie', 'Basketball', 'Bishop', 'Bronze', 'Crane', 'Kite', 'Sandpiper', 'Kangaroo', 'Daisy', 'Bat', 'Wasp', 'Subaru', 'Gold', 'Nut', 'Cardinal', 'Ford', 'Conch', 'Calf', 'Shoulder', 'Cream', 'Spring', 'Mortar', 'Pitcher', 'Cruiser', 'Mint', 'Nail', 'Referee', 'Cable car', 'Prickly pear', 'Iron', 'Silver', 'Crop', 'Foil', 'Hoe', 'Dairy', 'Punch', 'Hood', 'Fiat 500', 'Egg', 'Kung fu', 'Asparagus', 'Shortstraw pine', 'Propeller', 'Weasel', 'Windflower', 'Rose', 'Octopus', 'Dandelion', 'New york aster', 'Corn on the cob', 'Interior design', 'Volleyball', 'Trunk', 'Morning glory', 'Natal lily', 'Horn', 'Marble', 'Human leg', 'Wood', 'Greyhound', 'Sunflower', 'Lotus', 'Deck', 'Theatre', 'Peach', 'Audi r8', 'Armored car', 'Ivory', 'Course', 'Lavender', 'Jacuzzi', 'Jaguar', 'Tree frog', 'Kick', 'Chest', 'Turquoise', 'Lilac', 'Slate', 'Balsam fir', 'Samba', 'Swim brief', 'Lifejacket', 'Pear', 'Jumping', 'Birdhouse', 'Sailing', 'Cadillac', 'Sheep', 'Fish', 'Cup', 'Temple', 'Collar', 'Undergarment', 'Cricket', 'Fiat', 'Orange', 'Harrier', 'Shrub', 'Brine', 'Iris']
\n\nFor example:
\n\n'/m/02q2rdb' and '/m/05py0' both map to 'Octopus'. \n'/m/03v28z' and '/m/04v2k7' both map to 'Audi r8'
", "messages": [ { "id": 388934, "postDate": "2018-09-17T20:56:31.313Z", "content": "Each mid has a different meaning associated with it. The short descriptions are meant to be helpful in describing the mids, however the mids are the true labels that are to be used for evaluation. \nWhen raters verify the labels, they are given a more detailed description of the classes, beyond the single word description.
\n\nFor example:\n\"Trunk\" has 4 distinct mids referring to the trunk of a tree, the trunk of a car (also known as a boot in UK english), the trunk of a body (AKA torso), a traveling trunk.\nFor octopus, it looks like one mid refers to the family while the second refers to a particular species. \nThus, these are not duplicate label names, but rather mids with similar textual descriptions, which refer to different entities.
", "rawMarkdown": "Each mid has a different meaning associated with it. The short descriptions are meant to be helpful in describing the mids, however the mids are the true labels that are to be used for evaluation. \nWhen raters verify the labels, they are given a more detailed description of the classes, beyond the single word description.\n\nFor example:\n\"Trunk\" has 4 distinct mids referring to the trunk of a tree, the trunk of a car (also known as a boot in UK english), the trunk of a body (AKA torso), a traveling trunk.\nFor octopus, it looks like one mid refers to the family while the second refers to a particular species. \nThus, these are not duplicate label names, but rather mids with similar textual descriptions, which refer to different entities.\n\n", "votes": 2, "replies": [ { "id": 397323, "postDate": "2018-10-02T10:28:22.033Z", "content": "So it's a 19988-class task instead of 600-class classification task?
", "rawMarkdown": "So it's a 19988-class task instead of 600-class classification task?" }, { "id": 397342, "postDate": "2018-10-02T11:28:03.633Z", "content": "No. the trainable classes are provided in the data. its 7178 MIDs I believe. You can read more about this in the open images website on how the dataset is made. https://storage.googleapis.com/openimages/web/factsfigures.html
", "rawMarkdown": "No. the trainable classes are provided in the data. its 7178 MIDs I believe. You can read more about this in the open images website on how the dataset is made. https://storage.googleapis.com/openimages/web/factsfigures.html", "votes": 1 } ] }, { "id": 388131, "postDate": "2018-09-16T11:22:38.547Z", "content": "Hi,
\n\nThe following classes in the human labels have multiple keys that map to them. Is this an issue?. I guess predicting either of the keys then is counted as correct?
\n\n['Bun', 'String instrument', 'Organ', 'Cape', 'Ball', 'Runway', 'Dome', 'Zombie', 'Basketball', 'Bishop', 'Bronze', 'Crane', 'Kite', 'Sandpiper', 'Kangaroo', 'Daisy', 'Bat', 'Wasp', 'Subaru', 'Gold', 'Nut', 'Cardinal', 'Ford', 'Conch', 'Calf', 'Shoulder', 'Cream', 'Spring', 'Mortar', 'Pitcher', 'Cruiser', 'Mint', 'Nail', 'Referee', 'Cable car', 'Prickly pear', 'Iron', 'Silver', 'Crop', 'Foil', 'Hoe', 'Dairy', 'Punch', 'Hood', 'Fiat 500', 'Egg', 'Kung fu', 'Asparagus', 'Shortstraw pine', 'Propeller', 'Weasel', 'Windflower', 'Rose', 'Octopus', 'Dandelion', 'New york aster', 'Corn on the cob', 'Interior design', 'Volleyball', 'Trunk', 'Morning glory', 'Natal lily', 'Horn', 'Marble', 'Human leg', 'Wood', 'Greyhound', 'Sunflower', 'Lotus', 'Deck', 'Theatre', 'Peach', 'Audi r8', 'Armored car', 'Ivory', 'Course', 'Lavender', 'Jacuzzi', 'Jaguar', 'Tree frog', 'Kick', 'Chest', 'Turquoise', 'Lilac', 'Slate', 'Balsam fir', 'Samba', 'Swim brief', 'Lifejacket', 'Pear', 'Jumping', 'Birdhouse', 'Sailing', 'Cadillac', 'Sheep', 'Fish', 'Cup', 'Temple', 'Collar', 'Undergarment', 'Cricket', 'Fiat', 'Orange', 'Harrier', 'Shrub', 'Brine', 'Iris']
\n\nFor example:
\n\n'/m/02q2rdb' and '/m/05py0' both map to 'Octopus'. \n'/m/03v28z' and '/m/04v2k7' both map to 'Audi r8'
", "rawMarkdown": "Hi,\n\nThe following classes in the human labels have multiple keys that map to them. Is this an issue?. I guess predicting either of the keys then is counted as correct?\n\n['Bun', 'String instrument', 'Organ', 'Cape', 'Ball', 'Runway', 'Dome', 'Zombie', 'Basketball', 'Bishop', 'Bronze', 'Crane', 'Kite', 'Sandpiper', 'Kangaroo', 'Daisy', 'Bat', 'Wasp', 'Subaru', 'Gold', 'Nut', 'Cardinal', 'Ford', 'Conch', 'Calf', 'Shoulder', 'Cream', 'Spring', 'Mortar', 'Pitcher', 'Cruiser', 'Mint', 'Nail', 'Referee', 'Cable car', 'Prickly pear', 'Iron', 'Silver', 'Crop', 'Foil', 'Hoe', 'Dairy', 'Punch', 'Hood', 'Fiat 500', 'Egg', 'Kung fu', 'Asparagus', 'Shortstraw pine', 'Propeller', 'Weasel', 'Windflower', 'Rose', 'Octopus', 'Dandelion', 'New york aster', 'Corn on the cob', 'Interior design', 'Volleyball', 'Trunk', 'Morning glory', 'Natal lily', 'Horn', 'Marble', 'Human leg', 'Wood', 'Greyhound', 'Sunflower', 'Lotus', 'Deck', 'Theatre', 'Peach', 'Audi r8', 'Armored car', 'Ivory', 'Course', 'Lavender', 'Jacuzzi', 'Jaguar', 'Tree frog', 'Kick', 'Chest', 'Turquoise', 'Lilac', 'Slate', 'Balsam fir', 'Samba', 'Swim brief', 'Lifejacket', 'Pear', 'Jumping', 'Birdhouse', 'Sailing', 'Cadillac', 'Sheep', 'Fish', 'Cup', 'Temple', 'Collar', 'Undergarment', 'Cricket', 'Fiat', 'Orange', 'Harrier', 'Shrub', 'Brine', 'Iris']\n\nFor example: \n\n'/m/02q2rdb' and '/m/05py0' both map to 'Octopus'. \n'/m/03v28z' and '/m/04v2k7' both map to 'Audi r8'", "votes": 2 }, { "id": 397322, "postDate": "2018-10-02T10:28:06.103Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 388934, "author_name": "Pallavi", "author_url": "", "post_date": "2018-09-17T20:56:31.313000", "content": "Each mid has a different meaning associated with it. The short descriptions are meant to be helpful in describing the mids, however the mids are the true labels that are to be used for evaluation. \nWhen raters verify the labels, they are given a more detailed description of the classes, beyond the single word description.
\n\nFor example:\n\"Trunk\" has 4 distinct mids referring to the trunk of a tree, the trunk of a car (also known as a boot in UK english), the trunk of a body (AKA torso), a traveling trunk.\nFor octopus, it looks like one mid refers to the family while the second refers to a particular species. \nThus, these are not duplicate label names, but rather mids with similar textual descriptions, which refer to different entities.
", "votes": 2, "replies": [ { "id": 397323, "author_name": "KaiHoo", "author_url": "", "post_date": "2018-10-02T10:28:22.033000", "content": "So it's a 19988-class task instead of 600-class classification task?
", "votes": 0, "replies": [] }, { "id": 397342, "author_name": "Amrit Krishnan", "author_url": "", "post_date": "2018-10-02T11:28:03.633000", "content": "No. the trainable classes are provided in the data. its 7178 MIDs I believe. You can read more about this in the open images website on how the dataset is made. https://storage.googleapis.com/openimages/web/factsfigures.html
", "votes": 1, "replies": [] } ] }, { "id": 397322, "author_name": "", "author_url": "", "post_date": "2018-10-02T10:28:06.103000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "388934": "Each mid has a different meaning associated with it. The short descriptions are meant to be helpful in describing the mids, however the mids are the true labels that are to be used for evaluation. \nWhen raters verify the labels, they are given a more detailed description of the classes, beyond the single word description.\n\nFor example:\n\"Trunk\" has 4 distinct mids referring to the trunk of a tree, the trunk of a car (also known as a boot in UK english), the trunk of a body (AKA torso), a traveling trunk.\nFor octopus, it looks like one mid refers to the family while the second refers to a particular species. \nThus, these are not duplicate label names, but rather mids with similar textual descriptions, which refer to different entities.\n\n", "388131": "Hi,\n\nThe following classes in the human labels have multiple keys that map to them. Is this an issue?. I guess predicting either of the keys then is counted as correct?\n\n['Bun', 'String instrument', 'Organ', 'Cape', 'Ball', 'Runway', 'Dome', 'Zombie', 'Basketball', 'Bishop', 'Bronze', 'Crane', 'Kite', 'Sandpiper', 'Kangaroo', 'Daisy', 'Bat', 'Wasp', 'Subaru', 'Gold', 'Nut', 'Cardinal', 'Ford', 'Conch', 'Calf', 'Shoulder', 'Cream', 'Spring', 'Mortar', 'Pitcher', 'Cruiser', 'Mint', 'Nail', 'Referee', 'Cable car', 'Prickly pear', 'Iron', 'Silver', 'Crop', 'Foil', 'Hoe', 'Dairy', 'Punch', 'Hood', 'Fiat 500', 'Egg', 'Kung fu', 'Asparagus', 'Shortstraw pine', 'Propeller', 'Weasel', 'Windflower', 'Rose', 'Octopus', 'Dandelion', 'New york aster', 'Corn on the cob', 'Interior design', 'Volleyball', 'Trunk', 'Morning glory', 'Natal lily', 'Horn', 'Marble', 'Human leg', 'Wood', 'Greyhound', 'Sunflower', 'Lotus', 'Deck', 'Theatre', 'Peach', 'Audi r8', 'Armored car', 'Ivory', 'Course', 'Lavender', 'Jacuzzi', 'Jaguar', 'Tree frog', 'Kick', 'Chest', 'Turquoise', 'Lilac', 'Slate', 'Balsam fir', 'Samba', 'Swim brief', 'Lifejacket', 'Pear', 'Jumping', 'Birdhouse', 'Sailing', 'Cadillac', 'Sheep', 'Fish', 'Cup', 'Temple', 'Collar', 'Undergarment', 'Cricket', 'Fiat', 'Orange', 'Harrier', 'Shrub', 'Brine', 'Iris']\n\nFor example: \n\n'/m/02q2rdb' and '/m/05py0' both map to 'Octopus'. \n'/m/03v28z' and '/m/04v2k7' both map to 'Audi r8'", "397322": "" } }