Welcome to the Lyft 3D Object Detection for Autonomous Vehicles competition.
\n\nIn this competition, we're using LIDAR and image information captured by moving vehicles to predict the bounding volumes, orientation, and classes of all of the objects in a given scene.
\n\nLyft provides an excellent devkit and data tutorial notebook here, that can help you familiarize yourself with the data and how it all connects together (check out our Data tab as well). It also shows you how to use visualization tools to better see what the bounding volumes represent.
\n\nThe metric is a 3D version of 2D mAP - with a couple of simple translations between the contexts. Please check out the Evaluation page in the Overview tab for more details.
\n\nGood luck!
", "messages": [ { "id": "625009", "postDate": "09/12/2019 16:01:30", "content": "Welcome to the Lyft 3D Object Detection for Autonomous Vehicles competition.
\n\nIn this competition, we're using LIDAR and image information captured by moving vehicles to predict the bounding volumes, orientation, and classes of all of the objects in a given scene.
\n\nLyft provides an excellent devkit and data tutorial notebook here, that can help you familiarize yourself with the data and how it all connects together (check out our Data tab as well). It also shows you how to use visualization tools to better see what the bounding volumes represent.
\n\nThe metric is a 3D version of 2D mAP - with a couple of simple translations between the contexts. Please check out the Evaluation page in the Overview tab for more details.
\n\nGood luck!
", "rawMarkdown": "Welcome to the Lyft 3D Object Detection for Autonomous Vehicles competition.\n\nIn this competition, we're using LIDAR and image information captured by moving vehicles to predict the bounding volumes, orientation, and classes of all of the objects in a given scene.\n\nLyft provides an excellent devkit and data tutorial notebook [here](https://github.com/lyft/nuscenes-devkit), that can help you familiarize yourself with the data and how it all connects together (check out our Data tab as well). It also shows you how to use visualization tools to better see what the bounding volumes represent.\n\nThe metric is a 3D version of 2D mAP - with a couple of simple translations between the contexts. Please check out the [Evaluation](https://www.kaggle.com/c/3d-object-detection-for-autonomous-vehicles/overview/evaluation) page in the Overview tab for more details.\n\nGood luck!", "votes": null }, { "id": "625013", "postDate": "09/12/2019 16:08:54", "content": "Good luck everyone!
", "rawMarkdown": "Good luck everyone!", "votes": null }, { "id": "625033", "postDate": "09/12/2019 16:33:12", "content": "It's very interesting competition and I really want to join,,,,, but only if I could find the way to get enough resources....!
", "rawMarkdown": "It's very interesting competition and I really want to join,,,,, but only if I could find the way to get enough resources....!", "votes": null }, { "id": "625071", "postDate": "09/12/2019 17:21:44", "content": "Good luck everyone for new area of 3D Models VR era...
", "rawMarkdown": "Good luck everyone for new area of 3D Models VR era...", "votes": null }, { "id": "625113", "postDate": "09/12/2019 18:05:00", "content": "I don't know what do you want us to do with 30hr GPU per week.\nis this a joke? I mean, is completely impossible hahaha 85GB of data and complex models.\nAt least you could give GCP credits... at the end, don't you want us to use GCP?
", "rawMarkdown": "I don't know what do you want us to do with 30hr GPU per week.\nis this a joke? I mean, is completely impossible hahaha 85GB of data and complex models.\nAt least you could give GCP credits... at the end, don't you want us to use GCP?", "votes": null }, { "id": "625747", "postDate": "09/13/2019 12:04:18", "content": "Do you plan to publish an example notebook with python implementation of the evaluation criteria?
\n\nThis is probably the most complex metric I have seen in kaggle competitions so far :)
", "rawMarkdown": "Do you plan to publish an example notebook with python implementation of the evaluation criteria?\n\nThis is probably the most complex metric I have seen in kaggle competitions so far :)", "votes": null }, { "id": "625806", "postDate": "09/13/2019 13:27:09", "content": "Looking forward to this competition. Expecting to learn a lot from fellow Kagglers. Good luck to all participants.
", "rawMarkdown": "Looking forward to this competition. Expecting to learn a lot from fellow Kagglers. Good luck to all participants.", "votes": null }, { "id": "627715", "postDate": "09/16/2019 09:41:10", "content": "There is a mAP evaluation implementation in the devkit here: https://github.com/lyft/nuscenes-devkit/blob/master/lyft_dataset_sdk/eval/detection/mAP_evaluation.py \nAlthough it seems that you have to average over the different thresholds by yourself.
", "rawMarkdown": "There is a mAP evaluation implementation in the devkit here: https://github.com/lyft/nuscenes-devkit/blob/master/lyft_dataset_sdk/eval/detection/mAP_evaluation.py \nAlthough it seems that you have to average over the different thresholds by yourself.", "votes": null }, { "id": "627817", "postDate": "09/16/2019 12:23:28", "content": "Thanks @gaborfodor and @shuairan - yep, that's the implementation we based on C# code on!
", "rawMarkdown": "Thanks @gaborfodor and @shuairan - yep, that's the implementation we based on C# code on!", "votes": null }, { "id": "628314", "postDate": "09/17/2019 06:44:44", "content": "The github code doesn't match the evaluation described in the competitition. Github implements a classic AP (with area under precision-recall curve), while in Kaggle just an accuracy/Critical Success Index is used. Which one is really used here ? Since it is said that confidence scores are not important, I guess it is the metric described here, but it has little to do with the github implementation then.
", "rawMarkdown": "The github code doesn't match the evaluation described in the competitition. Github implements a classic AP (with area under precision-recall curve), while in Kaggle just an accuracy/Critical Success Index is used. Which one is really used here ? Since it is said that confidence scores are not important, I guess it is the metric described here, but it has little to do with the github implementation then.", "votes": null }, { "id": "634434", "postDate": "09/26/2019 09:22:22", "content": "Best of luck !
", "rawMarkdown": "Best of luck !", "votes": null }, { "id": "635853", "postDate": "09/28/2019 10:59:09", "content": "Good luck to all 🎉 😊!
", "rawMarkdown": "Good luck to all 🎉 😊!", "votes": null }, { "id": "636141", "postDate": "09/28/2019 21:26:52", "content": "@philculliton evaluation page needs a correction:
\n\nIn the submission file format section:\n''''''\n97ce3ab08ccbc0baae0267cbf8d4da947e1f11ae1dbcb80c3f4408784cd9170c,1.0 2742.15 673.16 -18.65 1.834 4.609 1.648 2.619 car
\n\nindicates that sample 97ce3ab08ccbc0baae0267cbf8d4da947e1f11ae1dbcb80c3f4408784cd9170c has a bounding volume with a confidence of 0.5, center_x of 2742.15, center_y of 673.16, center_z of -18.65, width of 1.834, length of 4.609, height of 1.648, yaw of 2.619, and a class_name of car.\n''''''\nbold numbers should be same.
", "rawMarkdown": "philculliton evaluation page needs a correction:\n\nIn the submission file format section:\n''''''\n97ce3ab08ccbc0baae0267cbf8d4da947e1f11ae1dbcb80c3f4408784cd9170c,**1.0** 2742.15 673.16 -18.65 1.834 4.609 1.648 2.619 car\n\nindicates that sample 97ce3ab08ccbc0baae0267cbf8d4da947e1f11ae1dbcb80c3f4408784cd9170c has a bounding volume with a confidence of **0.5**, center_x of 2742.15, center_y of 673.16, center_z of -18.65, width of 1.834, length of 4.609, height of 1.648, yaw of 2.619, and a class_name of car.\n''''''\nbold numbers should be same.", "votes": null }, { "id": "636643", "postDate": "09/30/2019 01:10:37", "content": "Yes, seems like a typo, probably a modified copy from the data page: https://www.kaggle.com/c/3d-object-detection-for-autonomous-vehicles/data
", "rawMarkdown": "Yes, seems like a typo, probably a modified copy from the data page: https://www.kaggle.com/c/3d-object-detection-for-autonomous-vehicles/data", "votes": null }, { "id": "641655", "postDate": "10/04/2019 23:16:06", "content": "In general, when you compute mAP. Confidence scores are used only for sorting.
\n\nThey are not important in the sense that if you square all of them, take log or apply any other monotonically increasing function you will get exactly the same mAP score.
", "rawMarkdown": "In general, when you compute mAP. Confidence scores are used only for sorting. \n\nThey are not important in the sense that if you square all of them, take log or apply any other monotonically increasing function you will get exactly the same mAP score.", "votes": null }, { "id": "642416", "postDate": "10/06/2019 03:43:09", "content": "thanks
", "rawMarkdown": "thanks", "votes": null }, { "id": "644369", "postDate": "10/08/2019 17:41:03", "content": "hi,\nThe requirements for hardware are huge.\nLooking at the LB not many results.\nMetaphorical speaking you came into a poor country and ask them to provide the best \"Foie Gras\" recipe. \nSorry for the sarcasm, but looking at laptop and my lack of knowledge and then to requirements...\nThis looks more like a hiring process then a competition.\nSuggestion: bring additional resources to the table (60 hours/week) so people can try more.\nwith respect,\nAdrian
", "rawMarkdown": "hi,\nThe requirements for hardware are huge.\nLooking at the LB not many results.\nMetaphorical speaking you came into a poor country and ask them to provide the best **\"Foie Gras\"** recipe. \nSorry for the sarcasm, but looking at laptop and my lack of knowledge and then to requirements...\nThis looks more like a hiring process then a competition.\nSuggestion: bring additional resources to the table (60 hours/week) so people can try more.\nwith respect,\nAdrian", "votes": null } ], "comments": [ { "id": 625013, "author_name": "bejeweled", "author_url": "", "post_date": "09/12/2019 16:08:54", "content": "Good luck everyone!
", "votes": null, "replies": [] }, { "id": 625033, "author_name": "bamps53", "author_url": "", "post_date": "09/12/2019 16:33:12", "content": "It's very interesting competition and I really want to join,,,,, but only if I could find the way to get enough resources....!
", "votes": null, "replies": [] }, { "id": 625071, "author_name": "seshurajup", "author_url": "", "post_date": "09/12/2019 17:21:44", "content": "Good luck everyone for new area of 3D Models VR era...
", "votes": null, "replies": [] }, { "id": 625113, "author_name": "jesucristo", "author_url": "", "post_date": "09/12/2019 18:05:00", "content": "I don't know what do you want us to do with 30hr GPU per week.\nis this a joke? I mean, is completely impossible hahaha 85GB of data and complex models.\nAt least you could give GCP credits... at the end, don't you want us to use GCP?
", "votes": null, "replies": [] }, { "id": 625747, "author_name": "gaborfodor", "author_url": "", "post_date": "09/13/2019 12:04:18", "content": "Do you plan to publish an example notebook with python implementation of the evaluation criteria?
\n\nThis is probably the most complex metric I have seen in kaggle competitions so far :)
", "votes": null, "replies": [ { "id": 627715, "author_name": "shuairan", "author_url": "", "post_date": "09/16/2019 09:41:10", "content": "There is a mAP evaluation implementation in the devkit here: https://github.com/lyft/nuscenes-devkit/blob/master/lyft_dataset_sdk/eval/detection/mAP_evaluation.py \nAlthough it seems that you have to average over the different thresholds by yourself.
", "votes": null, "replies": [] }, { "id": 627817, "author_name": "philculliton", "author_url": "", "post_date": "09/16/2019 12:23:28", "content": "Thanks @gaborfodor and @shuairan - yep, that's the implementation we based on C# code on!
", "votes": null, "replies": [] }, { "id": 628314, "author_name": "thomasgilles", "author_url": "", "post_date": "09/17/2019 06:44:44", "content": "The github code doesn't match the evaluation described in the competitition. Github implements a classic AP (with area under precision-recall curve), while in Kaggle just an accuracy/Critical Success Index is used. Which one is really used here ? Since it is said that confidence scores are not important, I guess it is the metric described here, but it has little to do with the github implementation then.
", "votes": null, "replies": [] }, { "id": 641655, "author_name": "iglovikov", "author_url": "", "post_date": "10/04/2019 23:16:06", "content": "In general, when you compute mAP. Confidence scores are used only for sorting.
\n\nThey are not important in the sense that if you square all of them, take log or apply any other monotonically increasing function you will get exactly the same mAP score.
", "votes": null, "replies": [] } ] }, { "id": 625806, "author_name": "manojprabhaakr", "author_url": "", "post_date": "09/13/2019 13:27:09", "content": "Looking forward to this competition. Expecting to learn a lot from fellow Kagglers. Good luck to all participants.
", "votes": null, "replies": [] }, { "id": 634434, "author_name": "krishnakatyal", "author_url": "", "post_date": "09/26/2019 09:22:22", "content": "Best of luck !
", "votes": null, "replies": [] }, { "id": 635853, "author_name": "hsinwenchang", "author_url": "", "post_date": "09/28/2019 10:59:09", "content": "Good luck to all 🎉 😊!
", "votes": null, "replies": [] }, { "id": 636141, "author_name": "rishabhiitbhu", "author_url": "", "post_date": "09/28/2019 21:26:52", "content": "@philculliton evaluation page needs a correction:
\n\nIn the submission file format section:\n''''''\n97ce3ab08ccbc0baae0267cbf8d4da947e1f11ae1dbcb80c3f4408784cd9170c,1.0 2742.15 673.16 -18.65 1.834 4.609 1.648 2.619 car
\n\nindicates that sample 97ce3ab08ccbc0baae0267cbf8d4da947e1f11ae1dbcb80c3f4408784cd9170c has a bounding volume with a confidence of 0.5, center_x of 2742.15, center_y of 673.16, center_z of -18.65, width of 1.834, length of 4.609, height of 1.648, yaw of 2.619, and a class_name of car.\n''''''\nbold numbers should be same.
", "votes": null, "replies": [ { "id": 636643, "author_name": "sergeyserebryakov", "author_url": "", "post_date": "09/30/2019 01:10:37", "content": "Yes, seems like a typo, probably a modified copy from the data page: https://www.kaggle.com/c/3d-object-detection-for-autonomous-vehicles/data
", "votes": null, "replies": [] } ] }, { "id": 642416, "author_name": "gyujin0113", "author_url": "", "post_date": "10/06/2019 03:43:09", "content": "thanks
", "votes": null, "replies": [] }, { "id": 644369, "author_name": "zinovadr", "author_url": "", "post_date": "10/08/2019 17:41:03", "content": "hi,\nThe requirements for hardware are huge.\nLooking at the LB not many results.\nMetaphorical speaking you came into a poor country and ask them to provide the best \"Foie Gras\" recipe. \nSorry for the sarcasm, but looking at laptop and my lack of knowledge and then to requirements...\nThis looks more like a hiring process then a competition.\nSuggestion: bring additional resources to the table (60 hours/week) so people can try more.\nwith respect,\nAdrian
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "625009": "Welcome to the Lyft 3D Object Detection for Autonomous Vehicles competition.\n\nIn this competition, we're using LIDAR and image information captured by moving vehicles to predict the bounding volumes, orientation, and classes of all of the objects in a given scene.\n\nLyft provides an excellent devkit and data tutorial notebook [here](https://github.com/lyft/nuscenes-devkit), that can help you familiarize yourself with the data and how it all connects together (check out our Data tab as well). It also shows you how to use visualization tools to better see what the bounding volumes represent.\n\nThe metric is a 3D version of 2D mAP - with a couple of simple translations between the contexts. Please check out the [Evaluation](https://www.kaggle.com/c/3d-object-detection-for-autonomous-vehicles/overview/evaluation) page in the Overview tab for more details.\n\nGood luck!", "625013": "Good luck everyone!", "625033": "It's very interesting competition and I really want to join,,,,, but only if I could find the way to get enough resources....!", "625071": "Good luck everyone for new area of 3D Models VR era...", "625113": "I don't know what do you want us to do with 30hr GPU per week.\nis this a joke? I mean, is completely impossible hahaha 85GB of data and complex models.\nAt least you could give GCP credits... at the end, don't you want us to use GCP?", "625747": "Do you plan to publish an example notebook with python implementation of the evaluation criteria?\n\nThis is probably the most complex metric I have seen in kaggle competitions so far :)", "625806": "Looking forward to this competition. Expecting to learn a lot from fellow Kagglers. Good luck to all participants.", "627715": "There is a mAP evaluation implementation in the devkit here: https://github.com/lyft/nuscenes-devkit/blob/master/lyft_dataset_sdk/eval/detection/mAP_evaluation.py \nAlthough it seems that you have to average over the different thresholds by yourself.", "627817": "Thanks @gaborfodor and @shuairan - yep, that's the implementation we based on C# code on!", "628314": "The github code doesn't match the evaluation described in the competitition. Github implements a classic AP (with area under precision-recall curve), while in Kaggle just an accuracy/Critical Success Index is used. Which one is really used here ? Since it is said that confidence scores are not important, I guess it is the metric described here, but it has little to do with the github implementation then.", "634434": "Best of luck !", "635853": "Good luck to all 🎉 😊!", "636141": "philculliton evaluation page needs a correction:\n\nIn the submission file format section:\n''''''\n97ce3ab08ccbc0baae0267cbf8d4da947e1f11ae1dbcb80c3f4408784cd9170c,**1.0** 2742.15 673.16 -18.65 1.834 4.609 1.648 2.619 car\n\nindicates that sample 97ce3ab08ccbc0baae0267cbf8d4da947e1f11ae1dbcb80c3f4408784cd9170c has a bounding volume with a confidence of **0.5**, center_x of 2742.15, center_y of 673.16, center_z of -18.65, width of 1.834, length of 4.609, height of 1.648, yaw of 2.619, and a class_name of car.\n''''''\nbold numbers should be same.", "636643": "Yes, seems like a typo, probably a modified copy from the data page: https://www.kaggle.com/c/3d-object-detection-for-autonomous-vehicles/data", "641655": "In general, when you compute mAP. Confidence scores are used only for sorting. \n\nThey are not important in the sense that if you square all of them, take log or apply any other monotonically increasing function you will get exactly the same mAP score.", "642416": "thanks", "644369": "hi,\nThe requirements for hardware are huge.\nLooking at the LB not many results.\nMetaphorical speaking you came into a poor country and ask them to provide the best **\"Foie Gras\"** recipe. \nSorry for the sarcasm, but looking at laptop and my lack of knowledge and then to requirements...\nThis looks more like a hiring process then a competition.\nSuggestion: bring additional resources to the table (60 hours/week) so people can try more.\nwith respect,\nAdrian" }, "source": "meta" }