Development of intelligent transportation systems (ITSs) has been an area of major focus of the freight industry in recent years. A significant amount of research has been done to enhance the perception of autonomous cars in order to reach a higher level of autonomy. With the development of vehicular communications, cooperative perception by sharing low-level data has become a viable approach to improving performance. In this work, cooperative object detection with point clouds is studied in depth. A framework based on the CARLA simulator is designed for fully simulated experiments, including scenario construction, data generation, deep learning model training, and evaluation. The novel composition of the training data produces a more generalized model. A fusion scheme that balances communication cost and performance is proposed. The testing results are evaluated and analyzed to learn the factors that decide the benefits of raw data sharing.