3D Reconstruction from Images: From Everyday Office Scenes to the Depths of the Human Body

Date:

Duration: 20 min

Location: General talk at CVLab, EPFL, Lausanne, Switzerland

Reconstructing 3D shapes from images is a fundamental challenge in computer vision, with wide-ranging applications from entertainment to scientific research. In this talk, I will share insights from three distinct use cases I’ve worked on, each leveraging different techniques to solve this problem. These techniques span traditional, hybrid, and cutting-edge methodologies, demonstrating their versatility across various real-world applications.

Inside the human body, LightNeuS is used to build a 3D model from images captured by an endoscope, taking advantage of the light source’s illumination decline as the camera moves. This allows for detailed 3D reconstructions similar to what could be obtained from expensive imaging techniques like CT or MRI scans. The key advantage is that LightNeuS provides not only a cost-effective solution but also a clear and detailed view of potential cancer lesions, improving early detection and treatment planning.

From our workspaces to the roads we drive on and even inside our bodies, the task of reconstructing 3D shapes from images is shaping the future and the possibilities for improving our everyday life continue to expand.