Obstacle Detection With StixelNet

Dependencies

• tested on the following environment:
  • OS: tested on Ubuntu 18.04
  • Tensorflow 2.0.0
  • python 3.6
• installing the dependencies:
  • python3 -m pip install -r requirements.txt

Training Data

Kitti Raw Dataset

Ground Truth

• Sample of automatically generated ground truth from velodyne pointclouds

Downloading the customized dataset for this repository

    python3 ./scripts/download_kitti_stixels.py

the dataset is about 5.4G, so would take sometime until finishing downloading.

StixelNet Model

Training

After downloading the dataset, run

    python3 ./train.py

model weights will be saved into ./saved_models directory

Test one image

• Download pretrained model weights with

    python3 ./scripts/download_kitti_stixels_model_weights.py

• Test on an image

    python3 ./test_single_image.py --model_path [path/to/model/weights]

Sample Result

References

• StixelNet: A Deep Convolutional Network for Obstacle Detection and Road Segmentation
• Real-time category-based and general obstacle detection for autonomous driving