YOLOv3 Tiny trained to predict Football

• Object detection using YOLOv3 Tiny is implemented. The tiny version was chosen to keep the model light, thus providing higher Frames Per Second(FPS) on mid and low end devices.

• Complete model was trained on Google Colab

• Dataset was acquired from Open Images Dataset

• YOLOv3 Tiny was trained with AlexeyAB's darknet framework

• Here's a YouTube video demonstrating the working of my model

• Model info-

  • Iterations - 3000
  • mean Average Precision (mAP) - 72 %

• All predictions were done using yolov3-tiny-obj_last.weights

Predicted image-