Implicit Neural Representation of volumes for high-resolution space carving

Implicit Neural Representation of points has become an increasingly popular method for representing 3D objects in a continuous function that maps any point in 3D space to a feature vector. However, the representation has made it tough to capture high- frequency details in the geometry of the objects. One way to improve the representation of such shapes is by learning high-frequency details using Fourier Features. We demon- strate the effectiveness of this approach by training a neural network to reconstruct 3D objects, using implicit neural representation and Fourier features. Our results show that incorporating Fourier Features into Implicit Neural Representation improves the accuracy and quality of reconstructed 3D objects.