ResNet152

ResNet-152, also known as ResNet-152, is a deep convolutional neural network that belongs to the Residual Network (ResNet) family. It was introduced in 2015 by Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun to enable the training of very deep networks through residual learning, which uses skip connections to improve gradient flow during backpropagation.

Architecturally, ResNet-152 uses a bottleneck design and stacks 152 layers. The network begins with a 7×7 convolution with stride 2, followed by a 3×3 max pooling. It then consists of four stages of residual blocks: conv2_x with 3 blocks, conv3_x with 8 blocks, conv4_x with 36 blocks, and conv5_x with 3 blocks. Each bottleneck block contains a 1×1 convolution that reduces dimensionality, a 3×3 convolution, and a 1×1 convolution that restores dimensionality, with a shortcut connection added to the block’s output. When downsampling occurs, the shortcut path may use a 1×1 projection to match the dimensionality.

The network ends with global average pooling and a fully connected layer for 1000-class ImageNet classification. ResNet-152 contains roughly 60 million parameters and is computationally intensive, but it delivers strong accuracy and is widely used as a backbone for image classification and transfer learning tasks. It is part of a family that includes ResNet-50 and ResNet-101, with ResNet-152 representing one of the deepest commonly used variants.