Commonly used pixel/voxel/fixed basis discretization of images do not scale well with resolution and problem dimension, and a fundamentally new approach is needed for such applications. In this WP, we represent images with a DNN, which can adapt its finite representation to each image individually. This idea is currently investigated in the context of solving high-dimensional PDEs; see also WP 1.2, WP 3.1 and WP 3.2, for simulating physical problems.