![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F636669%2Fa2c2f099b662ec8a0fc84c49bcef89e3%2FScreenshot%20from%202020-09-12%2017-57-21.png?generation=1600013110903313&alt=media\")
My understanding of Lyft and dataset paper
\nThe second-largest ride-sharing company in the US after Uber. They operate in 644 cities in the US and 12 cities in Canada. Additionally, they do services such as Food delivery, Bicycle sharing system, and many more.
\nIn 2012, their vision started on building self-driving vehicles. In this journey, they started investing and had a partnership with multiple companies.
\nIn 2017, introduced a separate self-driving division called Lyft Level 5.
\nOver a year, the team with size 300+ Engineers and Researchers and introduced self-driving vehicles (Employee Pilot) on public roads in California.
\nThe same year, acquired “Blue Vision Labs”. One of the reasons why Blue Vision Labs’ technology is so cutting-edge is that it can crowdsource highly detailed 3D maps of entire cities using just car-mounted camera phones. These maps allow a car to understand exactly where it is, what’s around it, and what to do next, with centimeter-level accuracy.
\nLevel 5 team wants to build data-driven solutions to fulfil their vision on self-driving vehicles since they have access to data which they got from their millions of rides every day. Same time they want to make this data useful and provide access to the community.
\nAs all we know, in 2019 released dataset for “3D Object Detection”- detecting traffic agents around AV. They believed in Deep Learning models.
\n
Also, they are very clear about their current and future pipeline. As mentioned below. For safe and efficient operation on roads, an Autonomous Vehicle (AV) should not only understand the current state of the nearby agents but also proactively anticipate their future behaviour. So, the next step is predicting the motion of agents around AV. This year, released dataset for “Motion prediction”.
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F636669%2F1657b6137a582cd2f6476bdc4215159f%2FScreenshot%20from%202020-09-12%2019-29-50.png?generation=1600013202967013&alt=media\")
So here, I am taking the opportunity to summarise “Lyft Level 5 dataset paper”
\nAs we know perception is the first step in the modern self-driving pipeline. Much work needs to be done around data-driven motion prediction of traffic participants, trajectory planning, and simulation before self-driving vehicles can become a reality.
\nDatasets for developing these methods differ from those used for perception, in that they require large amounts of behavioural observations and interactions. These are obtained by combining the output of perception systems with an understanding of the environment in the form of a semantic map that contains priors over expected behaviour.
\nBroad availability of datasets for these down-stream tasks is much more limited and mostly available only to large-scale industrial efforts in the form of in-house collected data. This limits the progress within the computer vision and robotics communities to advance modern machine learning systems for these important tasks.
\nSo, Lyft Level 5 team decided to share(opensource) the largest and most detail dataset with the community to train/explore models for motion prediction.
\nHere are the details of related dataset and their comparison.
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F636669%2F03f46ba0b455d12857a82c950494d9ac%2FScreenshot%20from%202020-09-13%2008-38-13.png?generation=1600013342104904&alt=media\")
Lyft prediction dataset differs in three substantial ways:
\nInstead of focusing on a wide city area, They provide 1000 hours of data along a single route. This is motivated by the assumption that, particularly in ride-hailing applications, first applications of deploying self-driving fleets are more likely to occur along a few high-demand routes. This makes it possible to bound requirements and quantifies accident risk.
They are contributing higher-quality scene data by providing full perception output including bounding boxes, class probabilities instead of pure centroids. In addition, their semantic map is more complete: it counts more than 15,000 human annotations instead of only lane centres.
They provide a high-resolution aerial image of the area. This is motivated by the fact that much of the information encoded in the semantic map is implicitly accessible in the aerial form.
Overview of the dataset
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F636669%2Fef45916cd68eda4b9493e2f02dcdb125%2FScreenshot%20from%202020-09-13%2006-44-53.png?generation=1600013457463883&alt=media\")
Statistics on dataset
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F636669%2F13937304b02d7c0133f87ee8bc5c3710%2FScreenshot%20from%202020-09-13%2009-44-20.png?generation=1600013516638464&alt=media\")
Examples from the scenes in the dataset.
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F636669%2F7605084579312da09e1ddddcb08deed4%2FScreenshot%20from%202020-09-13%2011-23-32.png?generation=1600013580948942&alt=media\")
Yet to understand the semantic map and Aerial map. Will update soon.
\nMy understanding of Lyft and dataset paper
\nThe second-largest ride-sharing company in the US after Uber. They operate in 644 cities in the US and 12 cities in Canada. Additionally, they do services such as Food delivery, Bicycle sharing system, and many more.
\nIn 2012, their vision started on building self-driving vehicles. In this journey, they started investing and had a partnership with multiple companies.
\nIn 2017, introduced a separate self-driving division called Lyft Level 5.
\nOver a year, the team with size 300+ Engineers and Researchers and introduced self-driving vehicles (Employee Pilot) on public roads in California.
\nThe same year, acquired “Blue Vision Labs”. One of the reasons why Blue Vision Labs’ technology is so cutting-edge is that it can crowdsource highly detailed 3D maps of entire cities using just car-mounted camera phones. These maps allow a car to understand exactly where it is, what’s around it, and what to do next, with centimeter-level accuracy.
\nLevel 5 team wants to build data-driven solutions to fulfil their vision on self-driving vehicles since they have access to data which they got from their millions of rides every day. Same time they want to make this data useful and provide access to the community.
\nAs all we know, in 2019 released dataset for “3D Object Detection”- detecting traffic agents around AV. They believed in Deep Learning models.
\n
Also, they are very clear about their current and future pipeline. As mentioned below. For safe and efficient operation on roads, an Autonomous Vehicle (AV) should not only understand the current state of the nearby agents but also proactively anticipate their future behaviour. So, the next step is predicting the motion of agents around AV. This year, released dataset for “Motion prediction”.
\n
So here, I am taking the opportunity to summarise “Lyft Level 5 dataset paper”
\nAs we know perception is the first step in the modern self-driving pipeline. Much work needs to be done around data-driven motion prediction of traffic participants, trajectory planning, and simulation before self-driving vehicles can become a reality.
\nDatasets for developing these methods differ from those used for perception, in that they require large amounts of behavioural observations and interactions. These are obtained by combining the output of perception systems with an understanding of the environment in the form of a semantic map that contains priors over expected behaviour.
\nBroad availability of datasets for these down-stream tasks is much more limited and mostly available only to large-scale industrial efforts in the form of in-house collected data. This limits the progress within the computer vision and robotics communities to advance modern machine learning systems for these important tasks.
\nSo, Lyft Level 5 team decided to share(opensource) the largest and most detail dataset with the community to train/explore models for motion prediction.
\nHere are the details of related dataset and their comparison.
\n
Lyft prediction dataset differs in three substantial ways:
\nInstead of focusing on a wide city area, They provide 1000 hours of data along a single route. This is motivated by the assumption that, particularly in ride-hailing applications, first applications of deploying self-driving fleets are more likely to occur along a few high-demand routes. This makes it possible to bound requirements and quantifies accident risk.
They are contributing higher-quality scene data by providing full perception output including bounding boxes, class probabilities instead of pure centroids. In addition, their semantic map is more complete: it counts more than 15,000 human annotations instead of only lane centres.
They provide a high-resolution aerial image of the area. This is motivated by the fact that much of the information encoded in the semantic map is implicitly accessible in the aerial form.
Overview of the dataset
\n
Statistics on dataset
\n
Examples from the scenes in the dataset.
\n
Yet to understand the semantic map and Aerial map. Will update soon.
\n