oist / Usiigaci

Usiigaci: stain-free cell tracking in phase contrast microscopy enabled by supervised machine learning
MIT License
192 stars 68 forks source link

how to improve the accuracy of model #28

Closed w1n2 closed 1 year ago

w1n2 commented 1 year ago

I use own dataset(DIC microscope images) to runing train.py. I don't modify other parameters. And I found the loss cann't be reduced after about 0.9. Can you give me some help to reduce the loss. My DIC microscope images have lower contrast.