We propose FiberNet, a method to estimate in-vivo the cardiac fiber architecture of the human atria from multiple catheter recordings of the electrical activation. Cardiac fibers play a central rolein the electro-mechanical function of the heart, yet they aredifficult to determine in-vivo, and hence rarely truly patient-specificin existing cardiac models.FiberNet learns the fibers arrangement by solvingan inverse problem with physics-informed neural networks. The inverse problem amounts to identifyingthe conduction velocity tensor of a cardiac propagation modelfrom a set of sparse activation maps. The use of multiple mapsenables the simultaneous identification of all the componentsof the conduction velocity tensor, including the local fiber angle.We extensively test FiberNet on synthetic 2-D and 3-D examples, diffusion tensor fibers, and a patient-specific case. We show that 3 maps are sufficient to accurately capture the fibers, also in thepresence of noise. With fewer maps, the role of regularization becomesprominent. Moreover, we show that the fitted model can robustlyreproduce unseen activation maps. We envision that FiberNet will help the creation of patient-specific models for personalized medicine.The full code is available at http://github.com/fsahli/FiberNet.