The developments of communication technologies, Internet of Things (IoT), and Artificial Intelligence (AI) have significantly accelerated the advancement of Intelligent Transportation Systems (ITS) and Autonomous Driving (AD) in recent years. The exchange of sensed information by widely deployed radars, cameras, and other sensors on vehicles and roadside infrastructure can improve the traffic awareness of drivers and pedestrians. However, wireless data transmission in vehicular networks is challenged by highly dynamic path loss due to utilized frequency bands, weather conditions, traffic overheads, and geographical conditions. In this paper, we propose an Integrated Sensing and Communication System (ISAC) based path loss prediction approach to improve the knowledge of wireless data transmissions in vehicular networks, which utilizes multi-modal data collected by millimeter-wave (mmWave) radars, laser radars, and cameras to forecast the end-to-end path loss distribution. By leveraging a generative adversarial network for parameter initialization coupled with fine-tuning through supervised learning, the model's accuracy can be significantly improved. To increase the model's scalability, the effects of weather conditions, geographical conditions, traffic overheads, and frequency bands are all analyzed. According to the simulation results, our model achieves excellent accuracy with Mean Squared Error (MSE) of the predicted path loss distribution below $3e^{-3}$ across five different scenarios.