Hi,
Easymode is running quite well on our cluster and the installation was straightforward. I'm having a question about the pixel sizes and the training sets that were used to train the models because we are experiencing issues with picking on tomograms that were collected at lower mag (3.5A/px).
We are having good success picking mito and ribosomes at 10A/px binned tomogram for data collected at 1.5-2A/px but not at 3.5A/px. I think this may be due to the binned input tomogram being bigger when the starting pixel size is larger. Internally, I presume that the model uses a fixed image size as input and that the training set may not have had a wide range of scaling augmentation. Please correct me if i'm wrong as I didn't explore the code yet.
Is there a solution to that? Like tiling larger tomograms into smaller chunks to fit the training set sizes?
Thanks
Hi,
Easymode is running quite well on our cluster and the installation was straightforward. I'm having a question about the pixel sizes and the training sets that were used to train the models because we are experiencing issues with picking on tomograms that were collected at lower mag (3.5A/px).
We are having good success picking mito and ribosomes at 10A/px binned tomogram for data collected at 1.5-2A/px but not at 3.5A/px. I think this may be due to the binned input tomogram being bigger when the starting pixel size is larger. Internally, I presume that the model uses a fixed image size as input and that the training set may not have had a wide range of scaling augmentation. Please correct me if i'm wrong as I didn't explore the code yet.
Is there a solution to that? Like tiling larger tomograms into smaller chunks to fit the training set sizes?
Thanks