Jedi technique for reducing convolutional networks — pruning
You are faced with the task of object detection again. The priority is to enhance speed while maintaining acceptable accuracy. You take the YOLOv3 architecture and finetune it. Accuracy (mAp75) exceeds 0.95. However, the processing speed is still low. Damn. Today we will sidestep quantization. Instead, we will look at Model Pruning — trimming excessive parts of the network to speed up inference without losing accuracy. Visually — from where, how much, and how […]
