I really enjoy the visualisation that cryoCARE is providing. I am working mostly with FIB tomograms without pretilt correction (to keep coordinates compatible with STA approaches), meaning that most of the reconstruction volume is empty (= does not contain any sample).
I was thus thinking whether the denoising training might be more efficient if it is only trained on coordinates within the actual sample volume, by applying a mask during training volume extraction. Does this make sense in your mind?
I'd be happy to try to implement it, eg. by adding a list of masks to train_data_config.json and parsing them as a boundary condition for CryoCAREDataset.create_random_coords?
Hi,
I really enjoy the visualisation that cryoCARE is providing. I am working mostly with FIB tomograms without pretilt correction (to keep coordinates compatible with STA approaches), meaning that most of the reconstruction volume is empty (= does not contain any sample). I was thus thinking whether the denoising training might be more efficient if it is only trained on coordinates within the actual sample volume, by applying a mask during training volume extraction. Does this make sense in your mind?
I'd be happy to try to implement it, eg. by adding a list of masks to
train_data_config.json
and parsing them as a boundary condition forCryoCAREDataset.create_random_coords
?Best, Benedikt