Christian S. Perone
Christian S. Perone
We should consider sticking with Python 3.8 only and then later -> Python 3.9, as 3.10 is almost there and upstream is Python 3.8,
We need to have support for dataset downloading automatically from l5kit (as a Python module) and with proper versioning of datasets.
[Here](https://github.com/lyft/l5kit/blob/19b7ac8a35f50467ee8916b58e60298b5f5e7146/l5kit/l5kit/rasterization/rasterizer_builder.py#L144) and [here](https://github.com/lyft/l5kit/blob/19b7ac8a35f50467ee8916b58e60298b5f5e7146/l5kit/l5kit/rasterization/rasterizer_builder.py#L161) we have a fallback to hard coded matrices, a warning message should be added there when this happens.
Currently, if we run `pytest` from outside of the `l5kit/l5kit` folder, it throws an error regarding a missing zarr dataset. It would be nice to support running `pytest` either on...
Reposting here as suggested in (https://forum.nuscenes.org/t/nuplan-paper-comparison-table/797): Hello ! I was reading the nuPlan paper (https://arxiv.org/abs/2106.11810) where the Table 1 (shown below):  Is showing that the Lyft dataset doesn't have...
There is a very nice approach [here](https://github.com/ternaus/TernausNetV2) where they predict not only the pixel-level classes but the touching areas as well, so it can be used to improve the seed...
According to the PR (https://github.com/pytorch/pytorch/pull/16743), it is now possible to pin memory for custom datasets. We should definitely implement that in order to overlap GPU copies and computation.
There are many transformations, such as `transforms.Resample` and `transforms.ElasticTransform` that aren't documented (with the Sphinx format).
Feedback
Hey, I just did a [simple integration](http://blog.christianperone.com/2018/10/pytorch-1-0-tracing-jit-and-libtorch-c-api-to-integrate-pytorch-into-nodejs/) of libtorch as an addon for nodejs. The API worked very nice and I **really enjoyed the design**, it's very well done !...