280 - Custom object segmentation using StarDist library in python

Code generated in the video can be downloaded from here:
Train: github.com/bns...
Predict: github.com/bns...
All other code:
github.com/bns...
This video tutorial explains the process of training your own StarDist model for object segmentation. It walks you through the process of importing training images and corresponding masks, training a model, and segmenting (any size) images using the trained model.
Warning: You may find this approach more efficient compared to U-net or Mask-RCNN

[@willberger96]

Thank you for the intro to StarDist! A great tool for our data science toolbelt!

[@pranaynirapure5795]

Thank you for uploading such useful content. I am still curious about 42!

[@rahulds001]

Sir we are very grateful to you for bringing these topics. One thing I would like to bring into your notice that we the remote sensing students who deal with satellite imagery we have no idea how to create the patch and mask for U-net segmentation as everyone works with available dataset and satellite imageries are geoTiff files. It would be very beneficial if you come out with this topic and also taking a specific objective like segmenting the water bodies or landslides in an satellite imagery using Unet algorithm. Thank you

[@racos01]

hey if you're looking on how to create your dataset of masks e.g. for water bodies, you can look up on previous videos of his titled "labelling images for semantic segmentation using label studio" on creating patches he has also made a video called "python tips and tricks 5: patches" hope this helps

[@DigitalSreeni]

I’ve done videos on all the topics you suggested. Please search my channel. I will try to create more targeted playlists.

[@ajay0909]

video 208 is the one you are looking for

[@brunospfc8511]

Thank You professor, and thanks for using Jupyter notebooks too, great channel!

[@louis-mbras8986]

A pleasure to listen to you. Thanks !

[@fadossi]

Learning a lot here! Thanks 👏🏼

[@tensaenigussie6970]

Hi sir, first i would like to appreciate for your nice explanation in line with any of your tutorials to have a basic understanding of different varieties of deep learning algorithms. so i would like to ask to make video on " Multi task deep learning approach for semantic segmentation" or is there any video tutorial you would like to refer it. Once again , Thank you so much !!

[@ranimsaidi9564]

Hi, Can you please explain what is the purpose behin augmentation in this case?

[@kaaziwa]

Thank you for your excellent tutorial videos! I recently followed your tutorial and successfully trained a model using my images. However, I'm having trouble converting the output files to a format compatible with QuPath. I have the following files in my output folder: weights_last.h5, weights_best.h5, and config.json. Do you have any suggestions on how to proceed?

[@akiskailidis3183]

how can i get the diffrent images and the mask? when i download i only get 1 image. can you help me with that?

[@rakshitrohan4155]

Can you please tell how to save a stardist model so it can be used later, since you said you imported a model which you had trained for 150 iterations. Also, thanks for the video!

[@aggelosdidachos3073]

What if there are overlapping nuclei/cells in the mask we are using for training? Shouldn't we somehow be able to draw their contours (till the point that we can see them) and label them ? Or does the training concern solely the detection of what is a nucleus/cell and what not ? I was wondering whether segmentation could be further improved if it was somehow part of the training procedure.

[@emmanuelailok4291]

Hi! What is the purpose of the masks in the dataset used? Because my dataset doesn't have masks. I'm a high school student working on the SipaKmed cervical cancer dataset and I'm having issues segmenting the Whole slide images. I would really appreciate your help

[@BismaHaroon-hx3qo]

Hello, is there any way I can access your dataset (the 256x256 version). I'm trying to generate the patches but it's not working with those.

[@uqixovi]

I would like to use Radiomics data of brain tumor mass for machine learning. How can I use a mask file to extract grayscale data of the mass from MRI images?

[@AKASHKUMAR-jz9yv]

Sir I want to segment the microstructure of deformed grains using machine learning. Please help me. My criteria for segmentation is based on shape, size and pixel density of grains and kernel average misorientation.

[@a.akhrif1492]

Can this custom model be used in Qupath?

[@manuelflores8220]

Thanks for the videos you're updating... It's so useful for my research! I would appreciate it if you share the file of the code because I cannot find it in your GitHub repository

[@FMH201]

Thank you for your efforts! Is it possible to extract the count of the segmented parts and their area?

[@DigitalSreeni]

I have covered it in many of my videos - using regionprops

[@RafiKhalil]

Great video sir kindly share the google colab file of this video and previous video, which are not available on your GitHub repository.

[@DigitalSreeni]

They're shared now.

[@ajay0909]

Hi sir, I was trying to implement your video 204. Everything was fine in spyder IDE but when I tried it in COLAB with the larger dataset the images and masks were not aligning. I am clueless why this is happening.

[@DigitalSreeni]

This video may help: kzbin.info/www/bejne/jn_JYnSKh56Cic0

[@ajay0909]

@@DigitalSreeni sir, that was for me. Thanks a lot for saving my time. I owe you one for this.

[@m_awaispak]

Thank you so much for this video. but i am facing a problem when i run the cell five it gives an error "list index out of range" how to solve this. anyone please help.

[@DigitalSreeni]

Check your list length and adjust so you are referencing something that is out of your list.

[@m_awaispak]

@@DigitalSreeni thank you for the answer..