There are many multi-objective optimization problems (MOPs) in real life that contain a large number of decision variables, such as auto body parts design, financial investment, engineering design, adversarial textual attack and so on. These problems are known as large-scale multi-objective optimization problems (LSMOPs). Due to the curse of dimensionality, existing multi-objective evolutionary algorithm encounter difficulties in balancing convergence and diversity on LSMOPs. In this paper, a Population Cooperation based Particle Swarm Optimization algorithm (PCPSO) is proposed for tackling LSMOPs. To be specific, PCPSO is a two-stage optimizer with two key components: (1) In the first stage, an inter-population collaboration component named Auxiliary Population Cooperation (APC) is used to improve the convergence speed. (2) In the second stage, an intra-subpopulation collaboration component called SubPopulation Cooperation (SPC) is applied to balance convergence and diversity. Experimental results on benchmark problems with up to 5000 decision variables and 2, 3, 5, 10 objectives demonstrate that the proposed PCPSO achieves better performance than several state-of-the-art large-scale multi-objective evolutionary algorithms (LSMOEAs) on most test problems.