Learning Efficient Multi-Agent Cooperative Visual Exploration