Python is the number one language in scientific development and research. Lately, the number of Python users in the field of inverse problems has increased significantly. One case is the 5D reconstruction of biological samples, where we can quickly set up and solve optimization problems and sometimes even write code “as math”. The large Python community helps even more because it provides tools for rapidly visualizing highly multidimensional data.

The talk explores and shows how the data looks, how Python aids in writing code that helps conceptualize and solve the problem, how we can write efficient algorithms for optimization, and how we can visualize and compare this data.

In more detail, the data comprises two tensors, a 3D volumetric CMOS acquisition and a 4D single-pixel camera image with two feature vectors per pixel. This data must be “fused” to retrieve an underlying high-resolution 5D tensor. This can be thought of as a super-resolution task but with aid from both tensors. Then, this is written down as an optimization problem, with classes representing mathematical operators, and is speeded up using Pytorch tensors.