Training-YOLO

• TAO Toolkit Training Examples
• Open Source Examples
• Training
  • YoloV4

TAO Toolkit Training Examples

• TAO Toolkit provides some example jupyter notebooks describing how to train an object detector using TAO Toolkit. You can download the TAO toolkit docker container (see Quick Start Guide) and try those examples. You can find all the training details in the TAO Toolkit Object Detection doc.
• Supported Data Formats:
  • TAO Toolkit Object Detection doc: KITTI Format
  • TAO Toolkit Object Detection doc: COCO Format
  • tfrecords
    • tfrecord-viewer

Open Source Examples

• We will use all of these examples to get some ideas. As you will notice, none of them will be sufficient for Carolo Cup.
• Popular YOLO implementations:
  • https://github.com/pjreddie/darknet
  • https://pjreddie.com/darknet/yolo/
  • https://github.com/AlexeyAB/darknet
• YOLO Training tips:
  • https://github.com/AlexeyAB/darknet#how-to-improve-object-detection
  • https://github.com/AlexeyAB/darknet/wiki
  • https://github.com/ultralytics/yolov5/discussions/2787#discussioncomment-611368
• Train YOLO on Colab:
  • YoloV3: https://colab.research.google.com/github/ultralytics/yolov3/blob/master/tutorial.ipynb
  • YoloV5: https://colab.research.google.com/github/ultralytics/yolov5/blob/master/tutorial.ipynb
• This example works in a similar way (but not the same) as the system we will build. Please, also have a look at the colab training notebooks which contain some hints for training the models!
  • https://github.com/Alzaib/Traffic-Signs-Detection-Tensorflow-YOLOv3-YOLOv4
• Another very similar example: https://www.kaggle.com/code/valentynsichkar/traffic-signs-detection-by-yolo-v3-opencv-keras/notebook

Training

YoloV4

• step-by-step notebooks:
  • https://github.com/Alzaib/Traffic-Signs-Detection-Tensorflow-YOLOv3-YOLOv4/blob/main/colab%20training%20code/yolov4_tiny_signs.ipynb
  • https://colab.research.google.com/drive/1mzL6WyY9BRx4xX476eQdhKDnd_eixBlG#scrollTo=GNVU7eu9CQj3
• when to stop training: https://github.com/AlexeyAB/darknet#when-should-i-stop-training
• https://github.com/AlexeyAB/darknet/wiki/Train-Detector-on-MS-COCO-(trainvalno5k-2014)-dataset
• detailed training settings: https://github.com/AlexeyAB/darknet#how-to-train-to-detect-your-custom-objects
• tips to improve object detection:
  • https://github.com/AlexeyAB/darknet#how-to-improve-object-detection
  • https://blog.roboflow.com/yolov4-tactics/
  • https://github.com/ultralytics/yolov5/discussions/2787#discussioncomment-611368