![]() Conv63 and Conv31 can be better choices when less complex models are needed. ResNet8 gives a good balance of performance and receptive field size. r ASSIGNMENT_RADIUS, -assignment-radius ASSIGNMENT_RADIUSĬlick here for a description of the model architectures, training methods, and training radiusĬurrently, there are several model architectures available for use as the region classifier targets TARGETS path to file specifying target particle coordinates Path to file containing predicted particle coordinates A threshold can then be chosen to optimize the F1 score or for specific recall/precision levels on a heldout set of micrographs. The topaz precision_recall_curve command can facilitate this by reporting the precision-recall curve for a list of predicted particle coordinates and a list of known target coordinates. Particles extracted using Topaz still have scores associated with them and a final particle list should be determined by choosing particles above some score threshold. Choosing a final particle list threshold (topaz precision_recall_curve) ![]() In this case, predicted coordinates must be assigned to target coordinates which requires an additional distance threshold (-assignment-radius). The radius parameter can be tuned automatically given a set of known particle coordinates by finding the radius which maximizes the average precision score.
0 Comments
Leave a Reply. |