The paper introduces SqueezeNet, a small CNN architecture that achieves AlexNet-level accuracy on the ImageNet dataset with 50 times fewer parameters and a model size of less than 0.5MB. The authors propose three main design strategies to reduce parameter count while maintaining accuracy: using 1x1 filters instead of 3x3 filters, decreasing the number of input channels to 3x3 filters, and delaying downsampling to later layers. They also introduce the Fire module, which consists of a squeeze layer followed by an expand layer, to facilitate these strategies. SqueezeNet is evaluated and compared to other models, demonstrating its effectiveness in terms of accuracy and model size. The paper further explores the design space of CNN architectures, including microarchitectural and macroarchitectural aspects, to understand how different choices impact model performance.The paper introduces SqueezeNet, a small CNN architecture that achieves AlexNet-level accuracy on the ImageNet dataset with 50 times fewer parameters and a model size of less than 0.5MB. The authors propose three main design strategies to reduce parameter count while maintaining accuracy: using 1x1 filters instead of 3x3 filters, decreasing the number of input channels to 3x3 filters, and delaying downsampling to later layers. They also introduce the Fire module, which consists of a squeeze layer followed by an expand layer, to facilitate these strategies. SqueezeNet is evaluated and compared to other models, demonstrating its effectiveness in terms of accuracy and model size. The paper further explores the design space of CNN architectures, including microarchitectural and macroarchitectural aspects, to understand how different choices impact model performance.