Convolutional Neural Networks (CNNs) have turned out to be the most sought-after algorithm for implementing many computer-vision tasks. But its huge number of computations and memory bandwidth requirement led to the need of hardware accelerators like GPUs and FPGAs. Due to the flexibility and energy efficiency offered by FPGAs over the other platforms, CNN inference accelerators based on FPGAs have now become a hot research area, especially in embedded and mobile applications where the resources are limited. So, this paper abstracts a review on the various state-of-the-art FPGA-based CNN inference accelerators. Also, a comparison is performed based on their performance and resource utilization.