Quantization of U-Net Model for Self-Driving

In this paper, we study U-Net, a semantic segmentation model for recognizing lanes and objects in self-driving vehicles, and apply quantization, a method to maximize the efficiency and performance of the model. The U-Net model is one of the Semantic Segmentation models widely used for image classification and object detection in the field of deep learning. It is very important to understand the surrounding environment accurately and respond appropriately in autonomous driving, and for this, the U-Net model is often used. However, since the U-Net model is a large-scale model, it has the disadvantage of a slow calculation speed and a large amount of memory used. To solve this problem, we propose a method of reducing the inference time and memory usage of the U-Net model by applying quantization technology. Quantization technology quantizes and stores model parameters and uses quantized values for calculation. This can reduce inference time and memory usage while maintaining the performance of the U-Net model. In this study, the quantization technique is applied to the UNet model, and the effect of quantization on the performance of the model is evaluated by analyzing the results.