Closed mzur closed 1 year ago
This can be extended to a MAIA model export/import. Models can be trained onshore on a potent GPU, then exported to an offshore instance (like the Jetson) to perform only inference.
Another idea is training on still images, storing the model and then applying it for inference on a video.
I removed the implemented model storage from master
again because it's not of immediate use. This is continued in #38. The transfer and import stuff can continue here.
Model transfer and re-use is not straight forward. We'll implement UnKnoT first #57. Model storage could only be used to download the model and use it offline outside BIIGLE. But this has a very low priority.
Allow storage and reuse of trained Mask R-CNN models. Users should be able to apply their models to other volumes, too.