My understanding of the task is we are predicting 50 future time steps of a vehicles position. We can predict up to 3 different options and assign a confidence value to each path, but on the sample submissions I see 3 confidence values as expected but x, y coordinates reaching as high as 249.
\nReading into the file it seems we can assign one path for 00-49 and another for 100-149 and the third 200-249. Is this correct? I did not see this explicitly stated anywhere
", "messages": [ { "id": "985545", "postDate": "08/25/2020 19:53:37", "content": "My understanding of the task is we are predicting 50 future time steps of a vehicles position. We can predict up to 3 different options and assign a confidence value to each path, but on the sample submissions I see 3 confidence values as expected but x, y coordinates reaching as high as 249.
\nReading into the file it seems we can assign one path for 00-49 and another for 100-149 and the third 200-249. Is this correct? I did not see this explicitly stated anywhere
", "rawMarkdown": "My understanding of the task is we are predicting 50 future time steps of a vehicles position. We can predict up to 3 different options and assign a confidence value to each path, but on the sample submissions I see 3 confidence values as expected but x, y coordinates reaching as high as 249. \n\nReading into the file it seems we can assign one path for 00-49 and another for 100-149 and the third 200-249. Is this correct? I did not see this explicitly stated anywhere", "votes": null }, { "id": "986337", "postDate": "08/26/2020 11:46:16", "content": "Hi @ryches , you intuitions are indeed correct!
\nHowever, we strongly suggest you to use our csv_utils to write predictions for both the single and multi-modal cases. It will take care or producing a correctly formatted csv ready to be submitted :)
Small typo, should be 00-49 in your post.
Good point. edited the post now
", "rawMarkdown": "Good point. edited the post now", "votes": null }, { "id": "1019373", "postDate": "09/20/2020 11:40:24", "content": "So in this case, for multi modal, prediction output shape must be batch_size x mode x feature_size x 2 right ?
", "rawMarkdown": "So in this case, for multi modal, prediction output shape must be batch_size x mode x feature_size x 2 right ?", "votes": null }, { "id": "1020625", "postDate": "09/21/2020 10:03:02", "content": "correct! see also here for the exact definition and implementation
", "rawMarkdown": "correct! see also [here](https://github.com/lyft/l5kit/blob/2d7d7654c438f9cdf5d7d85af98ea3a4c96653cf/l5kit/l5kit/evaluation/csv_utils.py#L157) for the exact definition and implementation", "votes": null }, { "id": "1052048", "postDate": "10/17/2020 08:37:36", "content": "is it true now after l5kit1.1.0
\nif so I don't understand.
\nwhile training with mode of 2 the output is 202
Hi @ryches , you intuitions are indeed correct!
\nHowever, we strongly suggest you to use our csv_utils to write predictions for both the single and multi-modal cases. It will take care or producing a correctly formatted csv ready to be submitted :)
Small typo, should be 00-49 in your post.
Good point. edited the post now
", "votes": null, "replies": [] } ] }, { "id": 1019373, "author_name": "ixtiyor", "author_url": "", "post_date": "09/20/2020 11:40:24", "content": "So in this case, for multi modal, prediction output shape must be batch_size x mode x feature_size x 2 right ?
", "votes": null, "replies": [ { "id": 1020625, "author_name": "lucabergamini", "author_url": "", "post_date": "09/21/2020 10:03:02", "content": "correct! see also here for the exact definition and implementation
", "votes": null, "replies": [] } ] }, { "id": 1052048, "author_name": "morizin", "author_url": "", "post_date": "10/17/2020 08:37:36", "content": "is it true now after l5kit1.1.0
\nif so I don't understand.
\nwhile training with mode of 2 the output is 202