{ "id": 177406, "title": "Welcome from hosts!", "url": "/competitions/lyft-motion-prediction-autonomous-vehicles/discussion/177406", "author_name": "Vladimir Iglovikov", "post_date": "2020-08-25T19:21:16.168000", "votes": 81, "comment_count": 23, "views": 0, "content": "

Hello everyone!

\n

The Lyft Level 5 team is excited to kick off this challenge on motion prediction for autonomous vehicles!

\n

The prediction problem is a crucial part of the autonomous stack and is currently an open research problem. We’re excited to provide you with the data and resources to contribute to solving this problem and can’t wait to see what you come up with.

\n

Task

\n\n

For this competition, you need to predict how the traffic agents will move in the next 50 frames. We ask that you predict a few trajectories for every agent and provide a confidence score for each of them.

\n

This problem is new to Kaggle, and currently lacks attention from the academic community. Last year, Lyft had a 3D Object Detection challenge. It was also an unfamiliar challenge to Kagglers. The third-place winner had only one GPU, and he was able to perform better than people with better hardware. I think a similar story may happen this time. Your domain knowledge, intelligence, and creativity will be your ticket to the top of the leaderboard, even if you do not have 100,500 GPUs in your devbox.

\n

I have participated in many Machine Learning competitions, and typically, I know what the winning approach will be. Not this time. You could go with heuristics, Kalman Filters, Computer Vision, or any other method.

\n

Hardware

\n

You can train models and generate CSV with submissions in

\n\n

Sample submission

\n

In our L5kit, we provide a sample model that rasterized the dataset and fed it through a Resnet50 type network.

\n

Kaggle is famous for its computer vision experts. Feel free to explore this direction and build on top of it.

\n

$300 GCP credits

\n

We are going to release the sample submission as a Kaggle kernel. It will have an entry on the leaderboard. If your submission scores higher, feel free to fill this survey and request a coupon for $300 GCP credits. The deadline is October 30. One coupon per user. The number of coupons is limited.

\n

TPU

\n

We will have additional prizes for people who use TPUs in their solutions. Details coming soon. :)

\n

Hosts

\n

We have a team of hosts that for this challenge:

\n\n

We are happy to help you with the questions about the challenge.

", "messages": [ { "id": 985521, "postDate": "2020-08-25T19:21:16.170Z", "content": "

Hello everyone!

\n

The Lyft Level 5 team is excited to kick off this challenge on motion prediction for autonomous vehicles!

\n

The prediction problem is a crucial part of the autonomous stack and is currently an open research problem. We’re excited to provide you with the data and resources to contribute to solving this problem and can’t wait to see what you come up with.

\n

Task

\n\n

For this competition, you need to predict how the traffic agents will move in the next 50 frames. We ask that you predict a few trajectories for every agent and provide a confidence score for each of them.

\n

This problem is new to Kaggle, and currently lacks attention from the academic community. Last year, Lyft had a 3D Object Detection challenge. It was also an unfamiliar challenge to Kagglers. The third-place winner had only one GPU, and he was able to perform better than people with better hardware. I think a similar story may happen this time. Your domain knowledge, intelligence, and creativity will be your ticket to the top of the leaderboard, even if you do not have 100,500 GPUs in your devbox.

\n

I have participated in many Machine Learning competitions, and typically, I know what the winning approach will be. Not this time. You could go with heuristics, Kalman Filters, Computer Vision, or any other method.

\n

Hardware

\n

You can train models and generate CSV with submissions in

\n\n

Sample submission

\n

In our L5kit, we provide a sample model that rasterized the dataset and fed it through a Resnet50 type network.

\n

Kaggle is famous for its computer vision experts. Feel free to explore this direction and build on top of it.

\n

$300 GCP credits

\n

We are going to release the sample submission as a Kaggle kernel. It will have an entry on the leaderboard. If your submission scores higher, feel free to fill this survey and request a coupon for $300 GCP credits. The deadline is October 30. One coupon per user. The number of coupons is limited.

\n

TPU

\n

We will have additional prizes for people who use TPUs in their solutions. Details coming soon. :)

\n

Hosts

\n

We have a team of hosts that for this challenge:

\n\n

We are happy to help you with the questions about the challenge.

", "rawMarkdown": "Hello everyone!\n\n\nThe [Lyft Level 5](https://self-driving.lyft.com/level5/) team is excited to kick off this challenge on motion prediction for autonomous vehicles!\n\nThe prediction problem is a crucial part of the autonomous stack and is currently an open research problem. We’re excited to provide you with the data and resources to contribute to solving this problem and can’t wait to see what you come up with.\n\n## Task\n\n* You have an HD map. \n* You know where we are on the map.\n* You know current and past positions for cars, bicyclists, pedestrians around us.\n\nFor this competition, you need to predict how the traffic agents will move in the next 50 frames. We ask that you predict a few trajectories for every agent and provide a confidence score for each of them.\n\nThis problem is new to Kaggle, and currently lacks attention from the academic community. Last year, Lyft had a [3D Object Detection challenge](https://www.kaggle.com/c/3d-object-detection-for-autonomous-vehicles/). It was also an unfamiliar challenge to Kagglers. The [third-place winner had only one GPU](https://www.kaggle.com/c/3d-object-detection-for-autonomous-vehicles/discussion/133895), and he was able to perform better than people with better hardware. I think a similar story may happen this time. Your domain knowledge, intelligence, and creativity will be your ticket to the top of the leaderboard, even if you do not have 100,500 GPUs in your devbox.\n\nI have participated in [many Machine Learning competitions](http://ternaus.blog/interview/2018/08/30/ama.html), and typically, I know what the winning approach will be. Not this time. You could go with heuristics, Kalman Filters, Computer Vision, or any other method. \n\n## Hardware\n\nYou can train models and generate CSV with submissions in \n* Kaggle Kernels on CPU, GPU, TPU.\n* Your hardware.\n\n## Sample submission\n\nIn our [L5kit](https://github.com/lyft/l5kit/), we provide a sample model that rasterized the dataset and fed it through a Resnet50 type network.\n\nKaggle is famous for its computer vision experts. Feel free to explore this direction and build on top of it.\n\n## $300 GCP credits\n\nWe are going to release the [sample submission](https://www.kaggle.com/lucabergamini/lyft-baseline-09-02) as a Kaggle kernel. It will have an entry on the leaderboard. If your submission scores higher, feel free to fill [this](https://www.kaggle.com/GCP_Credits_Form-Lyft2020_082420) survey and request a coupon for $300 GCP credits. The deadline is October 30. One coupon per user. The number of coupons is limited.\n\n## TPU\n\nWe will have additional prizes for people who use TPUs in their solutions. Details coming soon. :)\n\n## Hosts\n\nWe have a team of hosts that for this challenge:\n\n\n* Peter Ondrushka @pondruska - Head of AV Research\n* Luca Bergamini @lucabergamini - Software Engineer Intern at Lyft Level 5\n* Vladimir Iglovikov @iglovikov - Software Engineer at Lyft Level 5. Kaggle Grandmaster\n\nWe are happy to help you with the questions about the challenge.\n\n\n", "votes": 80 }, { "id": 1046764, "postDate": "2020-10-12T01:46:30.670Z", "content": "

Thanks for starting this competition. I wonder which is the kernel for the sample submission from Lyft? And, what is the score? It is difficult to tell it from the leaderboard and all the notebooks.

", "rawMarkdown": "Thanks for starting this competition. I wonder which is the kernel for the sample submission from Lyft? And, what is the score? It is difficult to tell it from the leaderboard and all the notebooks.", "votes": 1, "replies": [ { "id": 1046793, "postDate": "2020-10-12T02:40:51.040Z", "content": "

The score is for sample submission is 82.

", "rawMarkdown": "The score is for sample submission is 82.", "votes": 1 }, { "id": 1046798, "postDate": "2020-10-12T02:45:25.703Z", "content": "

Could you please a detail about when and how we get GCP Credits

", "rawMarkdown": "Could you please a detail about when and how we get GCP Credits" } ] }, { "id": 995762, "postDate": "2020-09-02T18:44:17.807Z", "content": "

I'm intrigued!

", "rawMarkdown": "I'm intrigued!", "votes": 1 }, { "id": 992719, "postDate": "2020-08-31T11:42:49.483Z", "content": "

Thanks for the competition! I hope I can contribute to the discussion!

", "rawMarkdown": "Thanks for the competition! I hope I can contribute to the discussion!", "votes": 1 }, { "id": 991605, "postDate": "2020-08-30T14:24:38.557Z", "content": "

Hi,
\nCan we write research papers based on our submission notebooks if we win the competition? And what if we don't win?
\nThanks.
\n@iglovikov

", "rawMarkdown": "Hi,\nCan we write research papers based on our submission notebooks if we win the competition? And what if we don't win?\nThanks.\n@iglovikov ", "votes": 1, "replies": [ { "id": 993062, "postDate": "2020-08-31T16:24:48.877Z", "content": "

You can write a research paper on the dataset if you win, or if you do not win or in any other circumstances. :)

", "rawMarkdown": "You can write a research paper on the dataset if you win, or if you do not win or in any other circumstances. :)", "votes": 3 } ] }, { "id": 991095, "postDate": "2020-08-30T06:27:10.163Z", "content": "

This looks fascinating. Thank you for this!

", "rawMarkdown": "This looks fascinating. Thank you for this!", "votes": 1 }, { "id": 990577, "postDate": "2020-08-29T17:49:17.970Z", "content": "

This challenge seems quite fascinating !

", "rawMarkdown": "This challenge seems quite fascinating !", "votes": 1 }, { "id": 992452, "postDate": "2020-08-31T07:02:09.967Z", "content": "

Hey, when will host's sample submission be available? When can we start applying for GCP credits?

", "rawMarkdown": "Hey, when will host's sample submission be available? When can we start applying for GCP credits?", "votes": 2 }, { "id": 988061, "postDate": "2020-08-27T18:29:57.557Z", "content": "

This challenge looks super interesting. Thanks for sharing!

", "rawMarkdown": "This challenge looks super interesting. Thanks for sharing!", "votes": 2 }, { "id": 985828, "postDate": "2020-08-26T03:32:11.940Z", "content": "

Thanks! Here's the link to questions: https://www.kaggle.com/c/lyft-motion-prediction-autonomous-vehicles/discussion/177376

", "rawMarkdown": "Thanks! Here's the link to questions: https://www.kaggle.com/c/lyft-motion-prediction-autonomous-vehicles/discussion/177376", "votes": 2 }, { "id": 1054008, "postDate": "2020-10-19T15:24:02.367Z", "content": "

Thanks for the competition.. I hope that I will be able to contribute ..

", "rawMarkdown": "Thanks for the competition.. I hope that I will be able to contribute .." }, { "id": 1050068, "postDate": "2020-10-15T03:29:21.923Z", "content": "

A question about the format of this competition. This competition is a code competition. However, it also allows one to use private kaggle dataset when submit from kernel. So is it allowed to upload your local inference result csv as a private Kaggle dataset then in the submission notebook, you simply read and output that private dataset?
\nIf it is allowed, then it looks like the code competition requirement is not necessary.

", "rawMarkdown": "A question about the format of this competition. This competition is a code competition. However, it also allows one to use private kaggle dataset when submit from kernel. So is it allowed to upload your local inference result csv as a private Kaggle dataset then in the submission notebook, you simply read and output that private dataset?\nIf it is allowed, then it looks like the code competition requirement is not necessary." }, { "id": 1047903, "postDate": "2020-10-13T03:44:24.747Z", "content": "

Actually, this competition is Kernel Submission competition. Due to some internet issue which happened due to heavy rain, 5 times the inference of time 2 hrs each got failed at the end. then I tried to submit through Colab but insisted me while submitting that \"kernel should be submitted\" Could you please find a solution for this.
\n@iglovikov

", "rawMarkdown": "Actually, this competition is Kernel Submission competition. Due to some internet issue which happened due to heavy rain, 5 times the inference of time 2 hrs each got failed at the end. then I tried to submit through Colab but insisted me while submitting that \"kernel should be submitted\" Could you please find a solution for this.\n@iglovikov ", "replies": [ { "id": 1048176, "postDate": "2020-10-13T08:52:42.813Z", "content": "

I thought when you click \"Save Version\", you can select \"Save & Run All\". Then it will rerun the entire notebook on the server, and you don't need to wait online for the inference result.

", "rawMarkdown": "I thought when you click \"Save Version\", you can select \"Save & Run All\". Then it will rerun the entire notebook on the server, and you don't need to wait online for the inference result." } ] }, { "id": 1042575, "postDate": "2020-10-08T10:21:36.133Z", "content": "

I have given the form for GCP Credits.
\nBut couldn't know if they got applied or not. There wasn't any message after I filled and submitted is it like so or I have done wrong. If I applied once more will it get rejected

", "rawMarkdown": "I have given the form for GCP Credits.\nBut couldn't know if they got applied or not. There wasn't any message after I filled and submitted is it like so or I have done wrong. If I applied once more will it get rejected" }, { "id": 1007933, "postDate": "2020-09-12T16:26:05.350Z", "content": "

When will the GCP credits be delivered?

", "rawMarkdown": "When will the GCP credits be delivered?" }, { "id": 986651, "postDate": "2020-08-26T16:55:49.347Z", "rawMarkdown": "", "isDeleted": true }, { "id": 991760, "postDate": "2020-08-30T16:40:17.820Z", "content": "

Thank you for sharing! This is great!

", "rawMarkdown": "Thank you for sharing! This is great!", "votes": 1 }, { "id": 991231, "postDate": "2020-08-30T08:45:53.307Z", "content": "

Thanks for sharing!

", "rawMarkdown": "Thanks for sharing!", "votes": 1 }, { "id": 988512, "postDate": "2020-08-28T05:12:51.363Z", "content": "

Thanks for sharing

", "rawMarkdown": "Thanks for sharing", "votes": 1 }, { "id": 987861, "postDate": "2020-08-27T15:22:01.717Z", "content": "

Thanks for sharing.

", "rawMarkdown": "Thanks for sharing.", "votes": 1 } ], "comments": [ { "id": 1046764, "author_name": "Louis Yang", "author_url": "", "post_date": "2020-10-12T01:46:30.670000", "content": "

Thanks for starting this competition. I wonder which is the kernel for the sample submission from Lyft? And, what is the score? It is difficult to tell it from the leaderboard and all the notebooks.

", "votes": 1, "replies": [ { "id": 1046793, "author_name": "Vladimir Iglovikov", "author_url": "", "post_date": "2020-10-12T02:40:51.040000", "content": "

The score is for sample submission is 82.

", "votes": 1, "replies": [] }, { "id": 1046798, "author_name": "Mohammed Rizin V K", "author_url": "", "post_date": "2020-10-12T02:45:25.703000", "content": "

Could you please a detail about when and how we get GCP Credits

", "votes": 0, "replies": [] } ] }, { "id": 995762, "author_name": "Austin Jones", "author_url": "", "post_date": "2020-09-02T18:44:17.807000", "content": "

I'm intrigued!

", "votes": 1, "replies": [] }, { "id": 992719, "author_name": "Onat Yapici", "author_url": "", "post_date": "2020-08-31T11:42:49.483000", "content": "

Thanks for the competition! I hope I can contribute to the discussion!

", "votes": 1, "replies": [] }, { "id": 991605, "author_name": "Surya Pratap Singh", "author_url": "", "post_date": "2020-08-30T14:24:38.557000", "content": "

Hi,
\nCan we write research papers based on our submission notebooks if we win the competition? And what if we don't win?
\nThanks.
\n@iglovikov

", "votes": 1, "replies": [ { "id": 993062, "author_name": "Vladimir Iglovikov", "author_url": "", "post_date": "2020-08-31T16:24:48.877000", "content": "

You can write a research paper on the dataset if you win, or if you do not win or in any other circumstances. :)

", "votes": 3, "replies": [] } ] }, { "id": 991095, "author_name": "Joshua Bernardino", "author_url": "", "post_date": "2020-08-30T06:27:10.163000", "content": "

This looks fascinating. Thank you for this!

", "votes": 1, "replies": [] }, { "id": 990577, "author_name": "Krishna Pal Deora", "author_url": "", "post_date": "2020-08-29T17:49:17.970000", "content": "

This challenge seems quite fascinating !

", "votes": 1, "replies": [] }, { "id": 992452, "author_name": "Kumar Shubham", "author_url": "", "post_date": "2020-08-31T07:02:09.967000", "content": "

Hey, when will host's sample submission be available? When can we start applying for GCP credits?

", "votes": 2, "replies": [] }, { "id": 988061, "author_name": "Rohan Giriraj", "author_url": "", "post_date": "2020-08-27T18:29:57.557000", "content": "

This challenge looks super interesting. Thanks for sharing!

", "votes": 2, "replies": [] }, { "id": 985828, "author_name": "Kaushal Shah", "author_url": "", "post_date": "2020-08-26T03:32:11.940000", "content": "

Thanks! Here's the link to questions: https://www.kaggle.com/c/lyft-motion-prediction-autonomous-vehicles/discussion/177376

", "votes": 2, "replies": [] }, { "id": 1054008, "author_name": "Atharva Sonanis", "author_url": "", "post_date": "2020-10-19T15:24:02.367000", "content": "

Thanks for the competition.. I hope that I will be able to contribute ..

", "votes": 0, "replies": [] }, { "id": 1050068, "author_name": "Louis Yang", "author_url": "", "post_date": "2020-10-15T03:29:21.923000", "content": "

A question about the format of this competition. This competition is a code competition. However, it also allows one to use private kaggle dataset when submit from kernel. So is it allowed to upload your local inference result csv as a private Kaggle dataset then in the submission notebook, you simply read and output that private dataset?
\nIf it is allowed, then it looks like the code competition requirement is not necessary.

", "votes": 0, "replies": [] }, { "id": 1047903, "author_name": "Mohammed Rizin V K", "author_url": "", "post_date": "2020-10-13T03:44:24.747000", "content": "

Actually, this competition is Kernel Submission competition. Due to some internet issue which happened due to heavy rain, 5 times the inference of time 2 hrs each got failed at the end. then I tried to submit through Colab but insisted me while submitting that \"kernel should be submitted\" Could you please find a solution for this.
\n@iglovikov

", "votes": 0, "replies": [ { "id": 1048176, "author_name": "Louis Yang", "author_url": "", "post_date": "2020-10-13T08:52:42.813000", "content": "

I thought when you click \"Save Version\", you can select \"Save & Run All\". Then it will rerun the entire notebook on the server, and you don't need to wait online for the inference result.

", "votes": 0, "replies": [] } ] }, { "id": 1042575, "author_name": "Mohammed Rizin V K", "author_url": "", "post_date": "2020-10-08T10:21:36.133000", "content": "

I have given the form for GCP Credits.
\nBut couldn't know if they got applied or not. There wasn't any message after I filled and submitted is it like so or I have done wrong. If I applied once more will it get rejected

", "votes": 0, "replies": [] }, { "id": 1007933, "author_name": "Arthur Emirov", "author_url": "", "post_date": "2020-09-12T16:26:05.350000", "content": "

When will the GCP credits be delivered?

", "votes": 0, "replies": [] }, { "id": 986651, "author_name": "", "author_url": "", "post_date": "2020-08-26T16:55:49.347000", "content": "", "votes": 0, "replies": [] }, { "id": 991760, "author_name": "Marton", "author_url": "", "post_date": "2020-08-30T16:40:17.820000", "content": "

Thank you for sharing! This is great!

", "votes": 1, "replies": [] }, { "id": 991231, "author_name": "Heena Jain", "author_url": "", "post_date": "2020-08-30T08:45:53.307000", "content": "

Thanks for sharing!

", "votes": 1, "replies": [] }, { "id": 988512, "author_name": "Kapilkumargupta", "author_url": "", "post_date": "2020-08-28T05:12:51.363000", "content": "

Thanks for sharing

", "votes": 1, "replies": [] }, { "id": 987861, "author_name": "Bivek Subedi", "author_url": "", "post_date": "2020-08-27T15:22:01.717000", "content": "

Thanks for sharing.

", "votes": 1, "replies": [] } ], "raw_markdown_by_id": { "985521": "Hello everyone!\n\n\nThe [Lyft Level 5](https://self-driving.lyft.com/level5/) team is excited to kick off this challenge on motion prediction for autonomous vehicles!\n\nThe prediction problem is a crucial part of the autonomous stack and is currently an open research problem. We’re excited to provide you with the data and resources to contribute to solving this problem and can’t wait to see what you come up with.\n\n## Task\n\n* You have an HD map. \n* You know where we are on the map.\n* You know current and past positions for cars, bicyclists, pedestrians around us.\n\nFor this competition, you need to predict how the traffic agents will move in the next 50 frames. We ask that you predict a few trajectories for every agent and provide a confidence score for each of them.\n\nThis problem is new to Kaggle, and currently lacks attention from the academic community. Last year, Lyft had a [3D Object Detection challenge](https://www.kaggle.com/c/3d-object-detection-for-autonomous-vehicles/). It was also an unfamiliar challenge to Kagglers. The [third-place winner had only one GPU](https://www.kaggle.com/c/3d-object-detection-for-autonomous-vehicles/discussion/133895), and he was able to perform better than people with better hardware. I think a similar story may happen this time. Your domain knowledge, intelligence, and creativity will be your ticket to the top of the leaderboard, even if you do not have 100,500 GPUs in your devbox.\n\nI have participated in [many Machine Learning competitions](http://ternaus.blog/interview/2018/08/30/ama.html), and typically, I know what the winning approach will be. Not this time. You could go with heuristics, Kalman Filters, Computer Vision, or any other method. \n\n## Hardware\n\nYou can train models and generate CSV with submissions in \n* Kaggle Kernels on CPU, GPU, TPU.\n* Your hardware.\n\n## Sample submission\n\nIn our [L5kit](https://github.com/lyft/l5kit/), we provide a sample model that rasterized the dataset and fed it through a Resnet50 type network.\n\nKaggle is famous for its computer vision experts. Feel free to explore this direction and build on top of it.\n\n## $300 GCP credits\n\nWe are going to release the [sample submission](https://www.kaggle.com/lucabergamini/lyft-baseline-09-02) as a Kaggle kernel. It will have an entry on the leaderboard. If your submission scores higher, feel free to fill [this](https://www.kaggle.com/GCP_Credits_Form-Lyft2020_082420) survey and request a coupon for $300 GCP credits. The deadline is October 30. One coupon per user. The number of coupons is limited.\n\n## TPU\n\nWe will have additional prizes for people who use TPUs in their solutions. Details coming soon. :)\n\n## Hosts\n\nWe have a team of hosts that for this challenge:\n\n\n* Peter Ondrushka @pondruska - Head of AV Research\n* Luca Bergamini @lucabergamini - Software Engineer Intern at Lyft Level 5\n* Vladimir Iglovikov @iglovikov - Software Engineer at Lyft Level 5. Kaggle Grandmaster\n\nWe are happy to help you with the questions about the challenge.\n\n\n", "1046764": "Thanks for starting this competition. I wonder which is the kernel for the sample submission from Lyft? And, what is the score? It is difficult to tell it from the leaderboard and all the notebooks.", "995762": "I'm intrigued!", "992719": "Thanks for the competition! I hope I can contribute to the discussion!", "991605": "Hi,\nCan we write research papers based on our submission notebooks if we win the competition? And what if we don't win?\nThanks.\n@iglovikov ", "991095": "This looks fascinating. Thank you for this!", "990577": "This challenge seems quite fascinating !", "992452": "Hey, when will host's sample submission be available? When can we start applying for GCP credits?", "988061": "This challenge looks super interesting. Thanks for sharing!", "985828": "Thanks! Here's the link to questions: https://www.kaggle.com/c/lyft-motion-prediction-autonomous-vehicles/discussion/177376", "1054008": "Thanks for the competition.. I hope that I will be able to contribute ..", "1050068": "A question about the format of this competition. This competition is a code competition. However, it also allows one to use private kaggle dataset when submit from kernel. So is it allowed to upload your local inference result csv as a private Kaggle dataset then in the submission notebook, you simply read and output that private dataset?\nIf it is allowed, then it looks like the code competition requirement is not necessary.", "1047903": "Actually, this competition is Kernel Submission competition. Due to some internet issue which happened due to heavy rain, 5 times the inference of time 2 hrs each got failed at the end. then I tried to submit through Colab but insisted me while submitting that \"kernel should be submitted\" Could you please find a solution for this.\n@iglovikov ", "1042575": "I have given the form for GCP Credits.\nBut couldn't know if they got applied or not. There wasn't any message after I filled and submitted is it like so or I have done wrong. If I applied once more will it get rejected", "1007933": "When will the GCP credits be delivered?", "986651": "", "991760": "Thank you for sharing! This is great!", "991231": "Thanks for sharing!", "988512": "Thanks for sharing", "987861": "Thanks for sharing." } }