Figure 6.

Architecture of the CryoSegNet model. (A) The attention-gated U-Net to predict segmentation mask for a micrograph. The numbers in the top of the rectangular slices indicate the number of channels and in the bottom indicate the size of the output. The U-Net has five encoders, one bottleneck component and five decoders. The skip connection from each encoder to its corresponding decoder goes through an attention gated block. Each attention block for a decoder also takes an input from its previous decoder or the bottleneck component. The details of the attention block are illustrated at the middle top. (B) The SAM mask generator takes input from the output of the U-Net model and outputs bounding box coordinates and intersection over union score for each predicted protein particle in the micrograph. (C) The postprocessing module outputs the star file containing picked particles and processed output micrographs based on the thresholding criterion for each protein type.