The integration of ultrasound localization microscopy (ULM) in ultrasound imaging has enabled an unprecedented enhancement in resolution, offering insights into blood flow direction and velocity measurement. Despite its potential, ULM remains a complex and time-intensive procedure, continually improved through advancements in deep learning (DL) methods. Current DL techniques for micro-bubble (MB) super-localization encounter complexity stemming from the use of high-resolution images within their networks. Consequently, these convolutional neural networks (CNNs) incur longer execution times compared to traditional ULM, necessitating arbitrary filtering of their outcomes prior to integration with a subsequent tracking algorithm. To address these challenges, our study introduces a novel DL approach inspired by the success of single-molecule localization microscopy DL techniques. Our proposed 3D CNN, named 3DML-ResNet, enables fast and scalable super-localization while providing explicit estimations of the number of MBs in each frame.