|
Figure 1
Overview of InstaMap. Cryo-EM images yi (left) with annotated pose and imaging parameters (Ri, Ti, PSFi) are used for gradient-based learning. Instant-NGP is queried at the rotated, shifted and jittered grid. A pose-independent projection matrix maps the 3D grid indices to the 2D plane (middle). The electron-microscope effects are then applied to generate a noiseless projected image corresponding to the observation of the top view of the biomolecule (TRPV1 ion channel; right). The loss function is computed by summing the losses from the individual particles. |