injekim / darknet_for_colab

Darknet repository for training YOLOv4 on Google Colab with Custom Dataset

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