Overview
MobileNetV3 is a series of convolutional neural network architectures designed for mobile and edge devices, focusing on maximizing accuracy while minimizing computational cost and model size. Based on the 'Searching for MobileNetV3' paper, these models leverage a combination of hardware-aware network architecture search (NAS) and network design. The architecture includes inverted residual blocks with linear bottlenecks and squeeze-and-excitation modules for efficient feature extraction. MobileNetV3 comes in two variants: Large and Small, catering to different resource constraints. These models are implemented within PyTorch's torchvision library, offering pre-trained weights for easy integration. They facilitate tasks such as image classification, object detection, and semantic segmentation, with optimized performance on resource-constrained devices. The builders instantiate a MobileNetV3 model from torchvision.models.mobilenetv3.MobileNetV3 base class.