benjaminrwilson / range-view-3d-detection

MIT License
10 stars 0 forks source link
3d-object-detection range-view

What Matters in Range View 3D Object Detection

Paper: https://arxiv.org/abs/2407.16789

Datasets

You will need to download both of these datasets in their entirety. After downloading, run the converters in ./converters. This will output each dataset in the format that our codebase expects.

Environment

You will need to install conda and install our conda environment. Please run:

bash conda/install.sh

This will install the environment to run the codebase. Note: This environment has only been tested on Ubuntu 20.04 using A40 gpus.

Additionally, to use weighted NMS you will need to install: https://github.com/Abyssaledge/TorchEx.

Training Script

The entrypoint is found in scripts. For example, to train our SOTA comparison model for Argoverse 2, you would run:

bash train.sh rv-av2 4 20 1.