Open
Description
This is a development requirement documentation of support LiDAR RCNN, which we wish to collaborate with the community.
In this issue, we wish to welcome the community developers to participate in the development of LiDAR RCNN.
Steps
- First have a discussion with the mmdet3d team about the development plan. We can divide the full algorithms into multiple modules and add each module step by step in different PR.
- In each PR, add a sub-module of LiDAR RCNN and add unit tests to test your code. The sub-module in this task can include a data converter, a data pipeline, or a component of the framework.
- Benchmark the performance when all modules have been added. The benchmark can be divided into two stages: first for inference and then for training.
- Prepare and release the models.
Resources
Official code release at https://github.com/TuSimple/LiDAR_RCNN