In high energy physics experiments Particle Flow (PFlow) algorithms are designed to reach optimal calorimeter reconstruction and jet energy resolution. A computer vision approach to PFlow reconstruction using deep Neural Network techniques based on Convolutional layers (cPFlow) is proposed. The algorithm is trained to learn, from calorimeter and charged particle track images, to distinguish the calorimeter energy deposits from neutral and charged particles in a non-trivial context, where the energy originated by a $\pi^{+}$ and a $\pi^{0}$ is overlapping within calorimeter clusters. The performance of the cPFlow and a traditional parametrized PFlow (pPFlow) algorithm are compared. The cPFlow provides a precise reconstruction of the neutral and charged energy in the calorimeter and therefore outperform more traditional pPFlow algorithm both, in energy response and position resolution.