Abstract
Quantum Implicit Neural Representations (QINRs) include components forlearning and execution on gate-based quantum computers. While QINRs recentlyemerged as a promising new paradigm, many challenges concerning theirarchitecture and ansatz design, the utility of quantum-mechanical properties,training efficiency and the interplay with classical modules remain. This paperadvances the field by introducing a new type of QINR for 2D image and 3Dgeometric field learning, which we collectively refer to as Quantum VisualField (QVF). QVF encodes classical data into quantum statevectors using neuralamplitude encoding grounded in a learnable energy manifold, ensuring meaningfulHilbert space embeddings. Our ansatz follows a fully entangled design oflearnable parametrised quantum circuits, with quantum (unitary) operationsperformed in the real Hilbert space, resulting in numerically stable trainingwith fast convergence. QVF does not rely on classical post-processing -- incontrast to the previous QINR learning approach -- and directly employsprojective measurement to extract learned signals encoded in the ansatz.Experiments on a quantum hardware simulator demonstrate that QVF outperformsthe existing quantum approach and widely used classical foundational baselinesin terms of visual representation accuracy across various metrics and modelcharacteristics, such as learning of high-frequency details. We also showapplications of QVF in 2D and 3D field completion and 3D shape interpolation,highlighting its practical potential.