Atelier Hypernetwork for Self-Supervised CryoEM Volume Reconstruction

Atelier uses a transformer-based hypernetwork to amortize implicit neural representation (INR) fitting for cryoEM maps. Pretrained on 5,439 maps, it generates scale-agnostic, coordinate-conditioned reconstructions that maintain spatial consistency across different protein structures.