Custom Darknet for training YOLOv4 on Google Colab with custom dataset

Setup darknet environment in Colab Notebook

To enable GPU backend for your notebook: Runtime->Change runtime type->Hardware Accelerator->GPU

# run these command line from notebook cell

!git clone https://github.com/quangnhat185/darknet_for_colab.git
%cd darknet_for_colab
!make
!chmod +x ./darknet

Tuning parameters from Colab environment

Double click on yolov4_config.pyto edit model parameters. More details about the meaning of each parameter can be found here

Generate YOLOv4 config and test file

!python yolov4_setup.py

Train with YOLOv4

!./darknet detector train data/yolov4.data cfg/yolov4_custom_train.cfg {weights_path} -map

Predict with YOLOv4

Image (predicted image is saved at predictions.jpg:

%cp data/yolov4.data cfg/coco.data
!./darknet detect cfg/yolov4_custom_test.cfg {weights_path} {img_path}

Video:

usage: darknet_video.py [-h] -v VIDEO [-c CONFIG] -w WEIGHTS [-l LABEL]
                [-m META] [-o OUTPUT]

optional arguments:
  -h, --help            show this help message and exit
  -v VIDEO, --video VIDEO
                        Path to input video
  -c CONFIG, --config CONFIG
                        Path to yolo config file
  -w WEIGHTS, --weights WEIGHTS
                        Path to yolo weight
  -l LABEL, --label LABEL
                        Path to label file
  -m META, --meta META  Path to metaPath
  -o OUTPUT, --output OUTPUT
                        Path to output file

!python darknet_video.py -v {video path} -c cfg/yolov4_custom_test.cfg -w {weights_path}  -o output.mp4

Tutorial

YOLOv4 in Google Colab: Train your Custom Dataset (Traffic Signs) with ease

License

MIT License