I wonder how could we split our 'LIDAR-TOP' channel data into different views, such as 'FRONT' view and 'BACK' view. Since it seems that LyftSDK does not provide such a function like 'get_pointcloud_in_front_view', we may need to define it from scratch.
", "messages": [ { "id": "651290", "postDate": "10/17/2019 09:33:53", "content": "I wonder how could we split our 'LIDAR-TOP' channel data into different views, such as 'FRONT' view and 'BACK' view. Since it seems that LyftSDK does not provide such a function like 'get_pointcloud_in_front_view', we may need to define it from scratch.
", "rawMarkdown": "I wonder how could we split our 'LIDAR-TOP' channel data into different views, such as 'FRONT' view and 'BACK' view. Since it seems that LyftSDK does not provide such a function like 'get_pointcloud_in_front_view', we may need to define it from scratch.", "votes": null }, { "id": "651319", "postDate": "10/17/2019 10:24:55", "content": "Easiest method would be a definition of valid X and Y values for each view and to filter the point cloud accordingly.
", "rawMarkdown": "Easiest method would be a definition of valid X and Y values for each view and to filter the point cloud accordingly.", "votes": null }, { "id": "651332", "postDate": "10/17/2019 10:49:05", "content": "You can do so by projecting the lidar points in front cam's view and filtering out those points which don't lie in the front cam image (something like this)
", "rawMarkdown": "You can do so by projecting the lidar points in front cam's view and filtering out those points which don't lie in the front cam image (something like [this](https://github.com/lyft/nuscenes-devkit/blob/master/lyft_dataset_sdk/utils/geometry_utils.py#L62))", "votes": null }, { "id": "656253", "postDate": "10/24/2019 04:10:43", "content": "Good Idea. BTW, I noticed that you mentioned the angle transformation under Reference Model of Guido. I believe that I could ask you questions about Euler angles. I am confused if the angle in box annotation of training data is YAW angle. Here is the annotation of a car, and is the ang(162.29) YAW angle?\n\nlabel: nan, score: nan, xyz: [1048.16, 1691.81, -23.30], wlh: [2.00, 5.28, 1.73], rot axis: [0.00, 0.00, 1.00], ang(degrees): 162.29, ang(rad): 2.83, vel: nan, nan, nan, name: car, token: 2a03c42173cde85f5829995c5851cc81158351e276db493b96946882059a5875\n
hi, did you find the answer to this question?
", "rawMarkdown": "hi, did you find the answer to this question?", "votes": null }, { "id": "666456", "postDate": "11/06/2019 05:39:57", "content": "I believe what @ilu000 mentioned could be a way to split dataset to different views.
", "rawMarkdown": "I believe what @ilu000 mentioned could be a way to split dataset to different views.", "votes": null }, { "id": "666509", "postDate": "11/06/2019 06:51:06", "content": "Sorry for late reply, Yes it is yaw, when the bounding box is flat on ground.
", "rawMarkdown": "Sorry for late reply, Yes it is yaw, when the bounding box is flat on ground.", "votes": null } ], "comments": [ { "id": 651319, "author_name": "ilu000", "author_url": "", "post_date": "10/17/2019 10:24:55", "content": "Easiest method would be a definition of valid X and Y values for each view and to filter the point cloud accordingly.
", "votes": null, "replies": [] }, { "id": 651332, "author_name": "rishabhiitbhu", "author_url": "", "post_date": "10/17/2019 10:49:05", "content": "You can do so by projecting the lidar points in front cam's view and filtering out those points which don't lie in the front cam image (something like this)
", "votes": null, "replies": [ { "id": 656253, "author_name": "usherbob", "author_url": "", "post_date": "10/24/2019 04:10:43", "content": "Good Idea. BTW, I noticed that you mentioned the angle transformation under Reference Model of Guido. I believe that I could ask you questions about Euler angles. I am confused if the angle in box annotation of training data is YAW angle. Here is the annotation of a car, and is the ang(162.29) YAW angle?\n\nlabel: nan, score: nan, xyz: [1048.16, 1691.81, -23.30], wlh: [2.00, 5.28, 1.73], rot axis: [0.00, 0.00, 1.00], ang(degrees): 162.29, ang(rad): 2.83, vel: nan, nan, nan, name: car, token: 2a03c42173cde85f5829995c5851cc81158351e276db493b96946882059a5875\n
hi, did you find the answer to this question?
", "votes": null, "replies": [] }, { "id": 666456, "author_name": "usherbob", "author_url": "", "post_date": "11/06/2019 05:39:57", "content": "I believe what @ilu000 mentioned could be a way to split dataset to different views.
", "votes": null, "replies": [] }, { "id": 666509, "author_name": "rishabhiitbhu", "author_url": "", "post_date": "11/06/2019 06:51:06", "content": "Sorry for late reply, Yes it is yaw, when the bounding box is flat on ground.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "651290": "I wonder how could we split our 'LIDAR-TOP' channel data into different views, such as 'FRONT' view and 'BACK' view. Since it seems that LyftSDK does not provide such a function like 'get_pointcloud_in_front_view', we may need to define it from scratch.", "651319": "Easiest method would be a definition of valid X and Y values for each view and to filter the point cloud accordingly.", "651332": "You can do so by projecting the lidar points in front cam's view and filtering out those points which don't lie in the front cam image (something like [this](https://github.com/lyft/nuscenes-devkit/blob/master/lyft_dataset_sdk/utils/geometry_utils.py#L62))", "656253": "Good Idea. BTW, I noticed that you mentioned the angle transformation under Reference Model of Guido. I believe that I could ask you questions about Euler angles. I am confused if the angle in box annotation of training data is YAW angle. Here is the annotation of a car, and is the ang(162.29) YAW angle?\n``\nlabel: nan, score: nan, xyz: [1048.16, 1691.81, -23.30], wlh: [2.00, 5.28, 1.73], rot axis: [0.00, 0.00, 1.00], ang(degrees): 162.29, ang(rad): 2.83, vel: nan, nan, nan, name: car, token: 2a03c42173cde85f5829995c5851cc81158351e276db493b96946882059a5875\n``", "666237": "hi, did you find the answer to this question?", "666456": "I believe what @ilu000 mentioned could be a way to split dataset to different views.", "666509": "Sorry for late reply, Yes it is yaw, when the bounding box is flat on ground." }, "source": "meta" }