The paper introduces CSRNet, a deep learning model designed for crowd counting and density map generation in highly congested scenes. CSRNet consists of two main components: a convolutional neural network (CNN) for 2D feature extraction and a dilated CNN for the back-end, which uses dilated kernels to expand receptive fields and replace pooling operations. The model is easy to train due to its pure convolutional structure. CSRNet is evaluated on four datasets (ShanghaiTech, UCF_CC_50, WorldEXPO'10, and UCSD) and achieves state-of-the-art performance, outperforming previous methods by up to 47.3% in Mean Absolute Error (MAE) on the ShanghaiTech Part-B dataset. The model is also extended to vehicle counting on the TRANCS dataset, achieving a 15.4% lower MAE compared to the best existing approach. The paper discusses the limitations of previous multi-scale architectures and demonstrates the effectiveness of CSRNet through experimental results and ablation studies.The paper introduces CSRNet, a deep learning model designed for crowd counting and density map generation in highly congested scenes. CSRNet consists of two main components: a convolutional neural network (CNN) for 2D feature extraction and a dilated CNN for the back-end, which uses dilated kernels to expand receptive fields and replace pooling operations. The model is easy to train due to its pure convolutional structure. CSRNet is evaluated on four datasets (ShanghaiTech, UCF_CC_50, WorldEXPO'10, and UCSD) and achieves state-of-the-art performance, outperforming previous methods by up to 47.3% in Mean Absolute Error (MAE) on the ShanghaiTech Part-B dataset. The model is also extended to vehicle counting on the TRANCS dataset, achieving a 15.4% lower MAE compared to the best existing approach. The paper discusses the limitations of previous multi-scale architectures and demonstrates the effectiveness of CSRNet through experimental results and ablation studies.