Interpreting Deep Picker Train Results

Hi all,

I ran a Deep Picker train on about 5000 particles and I got the following results.

Comparing this to the T20S dataset results, I obviously am missing the plots in the training and validation losses graph. The accuracies also seem to be off but I’m not sure what to expect per se. I’m unsure how to interpret these results. Any suggestions?

The Deep Picker inputs are as follows

Thank you


Update: After running the Deeppicker with a few different parameters, I went ahead and did the Deep Picker Inference. It did not pick any particles. I have a feeling the model may not have worked correctly. I’m not sure what I am doing wrong here. Any suggestions?

hi vamsee, did you managed to sort it out? I am having the same problem.

Hi Gorka, no, I haven’t been able to get around to it. I’ve been using Topaz for now but I’d really like to see if I get better or comparable results with Deeppicker.

Hi All,

I have similar problem in Cryosparc 4.4.1. I am training on 500 micrographs, which should contain roughly 20-30k particles. The input particle set was using templates and cleaned up by 2D class averaging. The class averages are crisp, 6Å images with clear alpha-helices and holes in beta-propellers visible.

Here are my input parameters:

Here are the training statistics. Seems like neither over nor underfitting the model.

Then I use Deep Inference module and the picker ends up picking nonsense (first image is Deep Picker, second image is template based picker):



Do you have any advice on how to tune the training? What could am I doing wrong?