Shadows significantly impact computer vision tasks, particularly in outdoor environments. State-of-the-art shadow removal methods are typically too computationally intensive for real-time image processing on edge hardware. We propose ShadowRemovalNet, a novel method designed for real-time image processing on resource-constrained hardware. ShadowRemovalNet achieves significantly higher frame rates compared to existing methods, making it suitable for real-time computer vision pipelines like those used in field robotics. Beyond speed, ShadowRemovalNet offers advantages in efficiency and simplicity, as it does not require a separate shadow mask during inference. ShadowRemovalNet also addresses challenges associated with Generative Adversarial Networks (GANs) for shadow removal, including artefacts, inaccurate mask estimations, and inconsistent supervision between shadow and boundary pixels. To address these limitations, we introduce a novel loss function that substantially reduces shadow removal errors. ShadowRemovalNet's efficiency and straightforwardness make it a robust and effective solution for real-time shadow removal in outdoor robotics and edge computing applications.