![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1905996%2F5dd34f0309265f1efc56854fe36b3347%2FScreenshot%202020-08-25%20at%206.52.31%20PM.png?generation=1598361774846916&alt=media\")
I'm trying to understand the dataset from this Notebook and in section called \"Working with raw data\", some trajectory is being plotted. It's iterating over all the frames (from all the scenes I assume) and plotting the \"centroids\" in scatter plot. Here is what I got:
What does this mean? Also what are \"centroid\"?
\nThanks in advance!
\nP.S.: I haven't participated in previous competitions of this series so I'm new to the terminology used here.
", "messages": [ { "id": "985053", "postDate": "08/25/2020 13:25:36", "content": "I'm trying to understand the dataset from this Notebook and in section called \"Working with raw data\", some trajectory is being plotted. It's iterating over all the frames (from all the scenes I assume) and plotting the \"centroids\" in scatter plot. Here is what I got:
What does this mean? Also what are \"centroid\"?
\nThanks in advance!
\nP.S.: I haven't participated in previous competitions of this series so I'm new to the terminology used here.
", "rawMarkdown": "I'm trying to understand the dataset from this [Notebook](https://github.com/lyft/l5kit/blob/master/examples/visualisation/visualise_data.ipynb) and in section called \"Working with raw data\", some trajectory is being plotted. It's iterating over all the frames (from all the scenes I assume) and plotting the \"centroids\" in scatter plot. Here is what I got: \n\nWhat does this mean? Also what are \"centroid\"?\n\nThanks in advance!\n\nP.S.: I haven't participated in previous competitions of this series so I'm new to the terminology used here.", "votes": null }, { "id": "985069", "postDate": "08/25/2020 13:37:07", "content": "I think \"ego_translation\" is a representation of the motion from the ego perspective (i.e the non-AV perspective in my thoughts)
", "rawMarkdown": "I think \"ego_translation\" is a representation of the motion from the ego perspective (i.e the non-AV perspective in my thoughts)", "votes": null }, { "id": "985083", "postDate": "08/25/2020 13:45:27", "content": "On this plot, these are all of the (x,y) coordinates in the dataset. They recorded the scenes on the same route. I think this is the route between Lyft HQ and stationX. (You can see the route if you plot the semantic map as well)
\nAV - Autonomous vehicle. These are Lyft's vehicles with lots of sensors.
\nEGO Vehicle - You are working with this agent's data (past, and predict its future). We can use any of the agents as \"EGO\" (use its past/future trajectory for training)
\nAgent - All of the participants on the roads (cars, pedestrians, bicycles, etc)
\nCentroid - is the middle of the vehicles/agents.
\nExtent - The 3d size of the EGO/Agent (width, height, length)
\nEGO Translation - Its position and yaw converted, so we can use it for training.
You can find more details about the coordinate systems here
", "rawMarkdown": "On this plot, these are all of the (x,y) coordinates in the dataset. They recorded the scenes on the same route. I think this is the route between Lyft HQ and stationX. (You can see the route if you plot the semantic map as well)\n\n**AV** - Autonomous vehicle. These are Lyft's vehicles with lots of sensors.\n**EGO Vehicle** - You are working with this agent's data (past, and predict its future). We can use any of the agents as \"EGO\" (use its past/future trajectory for training)\n**Agent** - All of the participants on the roads (cars, pedestrians, bicycles, etc)\n**Centroid** - is the middle of the vehicles/agents. \n**Extent** - The 3d size of the EGO/Agent (width, height, length)\n**EGO Translation** - Its position and yaw converted, so we can use it for training.\n\nYou can find more details about the coordinate systems [here](https://github.com/lyft/l5kit/blob/master/coords_systems.md)", "votes": null }, { "id": "985182", "postDate": "08/25/2020 14:55:31", "content": "@pestipeti thanks man!
", "rawMarkdown": "pestipeti thanks man!", "votes": null }, { "id": "985292", "postDate": "08/25/2020 16:14:41", "content": "Hey, I was myself dubious about including that plot in the notebook for this competition scope, so you're right to be confused by it :)
\nAs already written above this is a plot of one (or more) Lyft AV XY coordinates as it drives on a route.
\nAs such, these are the coordinates (centroids, translations, a lot of names for the same thing in this context) of the AV only. You can find more details about the different arrays in the zarr here
For this competition, you will focus on agents (other entities in the frame) NOT on the AV. That example has been included mainly to show how to get data from the raw zarr.
\nIn fact, while iterating over an AgentDataset you won't get any rasterisations from the AV perspective, but only from the agents ones.
Yes now it's clear! Thanks! :)
", "rawMarkdown": "Yes now it's clear! Thanks! :)", "votes": null }, { "id": "985351", "postDate": "08/25/2020 17:13:54", "content": "EGO == self … egomotion typically refers to the 3D movement of a camera through its environment. For AV, the vehicle under control is self, so EGO-translation, rotation, motion etc refer to the movement/position of that vehicle and its sensor package relative to its environment.
", "rawMarkdown": "EGO == self ... egomotion typically refers to the 3D movement of a camera through its environment. For AV, the vehicle under control is self, so EGO-translation, rotation, motion etc refer to the movement/position of that vehicle and its sensor package relative to its environment.", "votes": null } ], "comments": [ { "id": 985069, "author_name": "nxrprime", "author_url": "", "post_date": "08/25/2020 13:37:07", "content": "I think \"ego_translation\" is a representation of the motion from the ego perspective (i.e the non-AV perspective in my thoughts)
", "votes": null, "replies": [] }, { "id": 985083, "author_name": "pestipeti", "author_url": "", "post_date": "08/25/2020 13:45:27", "content": "On this plot, these are all of the (x,y) coordinates in the dataset. They recorded the scenes on the same route. I think this is the route between Lyft HQ and stationX. (You can see the route if you plot the semantic map as well)
\nAV - Autonomous vehicle. These are Lyft's vehicles with lots of sensors.
\nEGO Vehicle - You are working with this agent's data (past, and predict its future). We can use any of the agents as \"EGO\" (use its past/future trajectory for training)
\nAgent - All of the participants on the roads (cars, pedestrians, bicycles, etc)
\nCentroid - is the middle of the vehicles/agents.
\nExtent - The 3d size of the EGO/Agent (width, height, length)
\nEGO Translation - Its position and yaw converted, so we can use it for training.
You can find more details about the coordinate systems here
", "votes": null, "replies": [ { "id": 985182, "author_name": "kaushal2896", "author_url": "", "post_date": "08/25/2020 14:55:31", "content": "@pestipeti thanks man!
", "votes": null, "replies": [] } ] }, { "id": 985292, "author_name": "lucabergamini", "author_url": "", "post_date": "08/25/2020 16:14:41", "content": "Hey, I was myself dubious about including that plot in the notebook for this competition scope, so you're right to be confused by it :)
\nAs already written above this is a plot of one (or more) Lyft AV XY coordinates as it drives on a route.
\nAs such, these are the coordinates (centroids, translations, a lot of names for the same thing in this context) of the AV only. You can find more details about the different arrays in the zarr here
For this competition, you will focus on agents (other entities in the frame) NOT on the AV. That example has been included mainly to show how to get data from the raw zarr.
\nIn fact, while iterating over an AgentDataset you won't get any rasterisations from the AV perspective, but only from the agents ones.
Yes now it's clear! Thanks! :)
", "votes": null, "replies": [] } ] }, { "id": 985351, "author_name": "rwightman", "author_url": "", "post_date": "08/25/2020 17:13:54", "content": "EGO == self … egomotion typically refers to the 3D movement of a camera through its environment. For AV, the vehicle under control is self, so EGO-translation, rotation, motion etc refer to the movement/position of that vehicle and its sensor package relative to its environment.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "985053": "I'm trying to understand the dataset from this [Notebook](https://github.com/lyft/l5kit/blob/master/examples/visualisation/visualise_data.ipynb) and in section called \"Working with raw data\", some trajectory is being plotted. It's iterating over all the frames (from all the scenes I assume) and plotting the \"centroids\" in scatter plot. Here is what I got: \n\nWhat does this mean? Also what are \"centroid\"?\n\nThanks in advance!\n\nP.S.: I haven't participated in previous competitions of this series so I'm new to the terminology used here.", "985069": "I think \"ego_translation\" is a representation of the motion from the ego perspective (i.e the non-AV perspective in my thoughts)", "985083": "On this plot, these are all of the (x,y) coordinates in the dataset. They recorded the scenes on the same route. I think this is the route between Lyft HQ and stationX. (You can see the route if you plot the semantic map as well)\n\n**AV** - Autonomous vehicle. These are Lyft's vehicles with lots of sensors.\n**EGO Vehicle** - You are working with this agent's data (past, and predict its future). We can use any of the agents as \"EGO\" (use its past/future trajectory for training)\n**Agent** - All of the participants on the roads (cars, pedestrians, bicycles, etc)\n**Centroid** - is the middle of the vehicles/agents. \n**Extent** - The 3d size of the EGO/Agent (width, height, length)\n**EGO Translation** - Its position and yaw converted, so we can use it for training.\n\nYou can find more details about the coordinate systems [here](https://github.com/lyft/l5kit/blob/master/coords_systems.md)", "985182": "pestipeti thanks man!", "985292": "Hey, I was myself dubious about including that plot in the notebook for this competition scope, so you're right to be confused by it :) \n\nAs already written above this is a plot of one (or more) Lyft AV XY coordinates as it drives on a route.\nAs such, these are the coordinates (centroids, translations, a lot of names for the same thing in this context) of the AV only. You can find more details about the different arrays in the zarr [here](https://github.com/lyft/l5kit/blob/master/data_format.md)\n\nFor this competition, you will focus on agents (other entities in the frame) NOT on the AV. That example has been included mainly to show how to get data from the raw zarr. \nIn fact, while iterating over an `AgentDataset` you won't get any rasterisations from the AV perspective, but only from the agents ones.", "985324": "Yes now it's clear! Thanks! :)", "985351": "EGO == self ... egomotion typically refers to the 3D movement of a camera through its environment. For AV, the vehicle under control is self, so EGO-translation, rotation, motion etc refer to the movement/position of that vehicle and its sensor package relative to its environment." }, "source": "meta" }