add field dependence to enable large-FOV imaging
-
Before training the field-dependent localization network, one needs to modify some parameters in the "parameter_setting_exp.py" according to specific context. This includes settings of image size, noise, FOV segmentaion, et al.
-
Then "training_fd1.py" can be run to train a netwrok, which usually takes days of time in our case. Additionally, "training_fd0.py" which ignores field dependence, can be run for comparison.
-
Finally, given image folder and the trained network, one can run "inference.py" implement localization frame by frame. The localization result is saved in a .csv file in the given image folder.
-
We resort to ThunderSTORM, a plug-in of ImageJ, to demonstrate of localization results.
3D super-resolved image of mitochondria with high fidelity (depth range: 0-4 um):
3D super-resolved image of microtubules with high fidelity (depth range: 0-4 um):
Publication: Xiao, Dafei, et al. "Large-FOV 3D localization microscopy by spatially variant point spread function generation." Science Advances 10.10 (2024): eadj3656.