Abstract: When dealing with constrained multi-objective optimization problems (CMOPs), it is crucial to balance convergence, feasibility, and diversity. Existing constrained multi-objective evolutionary algorithms (CMOEAs) often struggle to achieve such a balance, resulting in poor performance of the algorithms in handling optimization problems with complex feasible regions. To address this long-standing challenge, this paper proposes a novel evolutionary algorithm based on a two-stage and two-population collaborative optimization framework, termed CMOEA-DD. In the global exploration stage, the first population (Pop1) achieves an effective balance between objectives and constraints through a dynamic ranking strategy. Specifically, while guiding the population to stably converge towards the true Pareto front, this strategy deliberately retains infeasible individuals with excellent objective values, as such individuals can provide valuable directional information for subsequent feasible region exploration and further enhance the algorithm’s convergence speed. Meanwhile, the second population (Pop2) ignores constraint conditions entirely and aims to conduct comprehensive and extensive exploration of the entire solution space, thereby avoiding the risk of being trapped in local feasible regions. Critically, Pop2 guides Pop1 to safely traverse potential infeasible regions by sharing high-quality offspring information, which helps Pop1 escape local optima and discover more promising feasible sub-regions. In the local development stage, Pop2 gradually increases the preference degree of individuals for the constraint conditions. At the same time, it improves the distribution effect of Pop1 by providing feasible individuals with good diversity. A large number of experiments conducted on four well-known test suites demonstrate that CMOEA-DD is more competitive than seven representative Constrained Multi-Objective Evolutionary Algorithms.