Skip to content

Latest commit

 

History

History
25 lines (16 loc) · 1.24 KB

README.md

File metadata and controls

25 lines (16 loc) · 1.24 KB

LaserNet in Tensorflow 2.0

In this project I've done my best to implement the work from Uber ATG on LaserNet. LaserNet is a 3D object detector for autonomous driving, with 3 distinguishing factors: range-view, low-latency, and probabilistic detection. See here for the original paper. The original paper uses Uber's proprietary dataset, and overfits on the small KITTI dataset, so for this project I am using Waymo Open Dataset.

Instructions:

Run save_ds.py to save the dataset to disk from the cloud bucket and perform initial preprocessing Run shard_ds.py to split the dataset into shards. This is not strictly necessary but improves results by performing an out-of-memory shuffle Run train.py to train on the dataset

What works:

Basic implementation of the architecture for vehicles, pedestrians, and cyclists

What's still not done:

  • Parameterize the scripts (yeah I was lazy and used hard-coded paths)
  • Mean-shift cluster (coming soon)
  • Adaptive NMS
  • Multi-modal distriutions for vehicles
  • Integration with Waymo Dataset evaluation metrics (also coming soon)

Example Results

top_image_2