Meeting 30.04.20 - TobiasSchmidtDE/DeepL-MedicalImaging GitHub Wiki

What happened so far

fixed docker cuda issues
normalizing module - set on hold until we have:
- combined the datasets
- imlemented architectures for which it makes sense to try out different normalization
implemented all simple architectures: densenet, resnet, inceptionnet
added experiment logging to the pipeline
merging datasets:
- PLCO (On hold)
- ChestX-Ray14 NIH: Downloaded and rescaled. Still working on reformating labels CSV
- more details see [dataset description](Data Set & Preprocessing Methods)
added generator implementation

dataset: NaN encoding
which pathologies we are interested in? See list of available pathologies in [dataset description](Data Set & Preprocessing Methods):
ssh access - public key?

move models to lrz storage
merging datasets
- Focus on chestX-Ray14 NIH
- Maybe add PLCO (discuss which pathologies we are interested in)
integrate and document generator
state of the art architectures
reimplement generator for optional resizing
Nico: find a radiologist