search

HenriquesLab / ZeroCostDL4Mic

ZeroCostDL4Mic: A Google Colab based no-cost toolbox to explore Deep-Learning in Microscopy
MIT License
566 stars 130 forks source link

SEM images without pre/post processing of size 1024x1024 can be used to train the model and the number of images of size 1024x1024 required to train the model #254

Closed anirbannavalarch closed 1 year ago

anirbannavalarch commented 1 year ago

Hi, I am looking forward to re-train the StarDist 2D model and generate a new model to process SEM images for my research. Samples training images (SEM) are attached herewith and the sizes of each of the images are 1024x1024. Further, all the training images are 8 bit. I am looking forward to know if the images that are attached herewith can be used as is to train the model without any pre/post processing of the images. Moreover, I am also looking forward to know how many images that I will be requiring to train the model. Please explicitly let me know the number of images. I will be using Google Colab ZeroCostDL4Mic to create the model.

TI_1 TI_2 ppen.

Desktop :

Thanks for all your help. Regards, Anirban

guijacquemet commented 1 year ago

Hi @anirbannavalarch , It depends on what you want to segment. StarDist would likely do a very good job a segmented the ovale shape structures on your images. However StarDist will likely not perform well on the very elongated shapes here.

If you want to segment everything, I would recommend using the Unet notebook instead.

Best wishes Guillaume

anirbannavalarch commented 1 year ago

Hi Guillaume, Thanks for all your help and guidance. Highly appreciated! I will work on Unet notebook and use the same segmented training and mask images to create the tensor flow model and verify the model using separate quality control training source and mask image. However, I am not able to use the tensor flow model created in StarDist and U-net in ImageJ software as processing of section 5.3 is creating some errors. Please help me fix the error such that I can convert the tensor flow model into a compatible ImageJ plugin. If possible, please outline the necessary steps to fix the error. Regards, Anirban

On Sun, 30 Apr 2023 at 09:56, guijacquemet @.***> wrote:

Hi @anirbannavalarch https://github.com/anirbannavalarch , It depends on what you want to segment. StarDist would likely do a very good job a segmented the ovale shape structures on your images. However StarDist will likely not perform well on the very elongated shapes here.

If you want to segment everything, I would recommend using the Unet notebook instead.

Best wishes Guillaume

— Reply to this email directly, view it on GitHub https://github.com/HenriquesLab/ZeroCostDL4Mic/issues/254#issuecomment-1529043749, or unsubscribe https://github.com/notifications/unsubscribe-auth/A6SPULIHFCQFTH2EQNAN553XDZ4Q7ANCNFSM6AAAAAAWMF4ORQ . You are receiving this because you were mentioned.Message ID: @.***>