agrilive/yolov4-object-detection

An implementation of the AlexeyAB Darknet repo on a custom dataset

YOLOv4 Object Detection on Windows

Objective: To accurately detect the obstacles in the field so that our robot can avoid these obstacles.

Dataset: 53 images of obstacles in the field

Approach

Use AlexeyAB darknet implementation of Yolov4, which has support for Windows.

Steps

1. Set up GPU and darknet
   • Set up virtual environment
   • Install CUDA and cuDNN
   • Set up vcpkg library manager
   • Set up darknet on Windows
2. Generate augmented images
   • 4 augmentations per given image were generated
   • Augment bounding box in the image augmentation process
3. Train and test model
   • Follow darknet tutorial to train on custom dataset
   • Visualize predictions

The detailed steps can be found in the Jupyter Notebooks.