Scaled-YOLOv4: Scaling Cross Stage Partial Network
“The deep learning-based object detection technique has many applications in our daily life. For example, medical image analysis, self-driving vehicles, business analytics, and face identification all rely on object detection. The computing facilities required for the above applications maybe cloud computing facilities, general GPU, IoT clusters, or single embedded device. In order to design an effective object detector, model scaling technique is very important, because it can make object detector achieve high accuracy and real-time inference on various types of devices.
The most common model scaling technique is to change the depth (number of convolutional layers in a CNN) and width (number of convolutional filters in a convolutional layer) of the backbone, and then train CNNs suitable for different devices. For example among the ResNet series, ResNet-152 and ResNet-101 are often used in cloud server GPUs, ResNet-50 and ResNet-34 are often used in personal computer GPUs, and ResNet-18 and ResNet-10 can be used in low-end embedded systems…”