DenseNet121

DenseNet121 is a 121-layer Convolutional Neural Network (CNN) employing dense connectivity to maximize feature reuse and improve gradient flow: it significantly reduces the parameter count while maintaining high accuracy.

DenseNet121 is a highly efficient 121-layer deep learning architecture (a DenseNet-BC variant). Its core innovation is the dense block: each layer receives feature maps from all preceding layers via concatenation, ensuring maximum information flow and feature reuse. The network uses bottleneck layers (1x1 and 3x3 convolutions) and transition layers with compression to manage the feature map growth. This design effectively alleviates the vanishing-gradient problem and substantially reduces the number of parameters compared to traditional CNNs, achieving state-of-the-art results on benchmarks like ImageNet with greater computational efficiency.

https://github.com/liuzhuang13/DenseNet