The recent advancements in the field of computer vision and medical imaging have shown a significant shift towards leveraging deep learning techniques for real-time and highly accurate calibration and reconstruction tasks. Innovations in camera and projector calibration, as well as in cone beam computed tomography (CBCT) reconstruction, are particularly noteworthy. The integration of neural networks with traditional algorithms has led to methods that are not only more efficient but also more adaptable to dynamic environments. These developments are paving the way for more robust and flexible systems, particularly in applications requiring real-time recalibration and customized orbits for imaging. The field is also witnessing a trend towards open-source implementations, which facilitate broader adoption and further research. Notably, the use of differentiable models and shift-variant algorithms is emerging as a powerful approach to address the computational and memory challenges associated with traditional methods, leading to faster and more accurate results. These advancements are crucial for the advancement of interventional medical imaging and multi-camera systems, where real-time accuracy and adaptability are paramount.

Noteworthy Papers:

• A novel neural network-based method for real-time recalibration of infrared multi-camera systems demonstrates superior accuracy and adaptability.
• An innovative differentiable reconstruction method for arbitrary CBCT orbits significantly reduces computational time and improves image quality.