Deep neural networks have been utilized in a variety of applications and have shown to have exceptional skills in the area of computer vision. Complex network designs delivers a considerable computational resource and energy cost issue for real-time deployment. These difficulties can be solved using improvements like network compression. Many times, network compression may be achieved with minimum loss of accuracy. Accuracy may even enhance in rare circumstances. This study presents a pruning survey on network compression. Pruning can be classified as dynamic or static, depending on whether it is done offline or in real time. This article analyses pruning methods and explains the criterion for removing duplicate calculations. Also covered trade-offs in element-by-element, channel-by-channel, shape-by-shape, filter-by-filter, layer-by-layer, and even network-by-network pruning. In this article, the pros and limitations of a variety of existing methodologies are contrasted and analyzed, as well as compressed network accuracy findings for a variety of frameworks and practical advice for compressing networks.