Tutorial 118 - Binary semantic segmentation using U-Net (in Keras)

Поделиться
HTML-код
  • Опубликовано: 15 янв 2025

Комментарии • 44

  • @nibhanapuri4476
    @nibhanapuri4476 3 года назад +3

    Good morning,
    I working with diabetic retinopathy lesion detection using unet. But I am unable to produce good segmentation results .can you suggest me tips to improve the model and hyper parameter tunning

    • @antoniojunior4853
      @antoniojunior4853 3 года назад

      did you figure out any solution? I'm having the same problem.. I'm getting reasonable results of accuracy recall etc.. but Im not able to produce good segmentation...

    • @taharabs8006
      @taharabs8006 9 месяцев назад

      did u guys figure it out !!

  • @v_explore4123
    @v_explore4123 3 месяца назад

    Thank you so much sir! In your opinion, can you suggest if there is any CNN which is better than U-Net on segmentation for biomedical imaging ?

  • @蘇裕宸-j2m
    @蘇裕宸-j2m 2 года назад +2

    Sir,what should i do to get the confidence of the object?
    Thx you very much

  • @shynimoon123
    @shynimoon123 Год назад

    Dear Sreeni sir , this video helped me so much .. thanks a lot

  • @fernandocanizares9521
    @fernandocanizares9521 3 года назад +1

    Thx you very much for your example and your nice explanation. Hope to see more content from you!

  • @sohailmaqsood381
    @sohailmaqsood381 Год назад +2

    I am facing this error. Can somebody help me?
    image_dataset = np.array(images)
    image_dataset = np.expand_dims(image_dataset, axis = 3)
    AxisError: axis 3 is out of bounds for array of dimension 2

    • @jayeshpattamatta8958
      @jayeshpattamatta8958 Год назад

      I am also facing the same error if I keep this axis as 3.
      mask_dataset = np.array(masks)
      ValueError: could not broadcast input array from shape (1850,1748,3) into shape (1850,) is coming due to this

  • @nataliameira2283
    @nataliameira2283 Год назад

    What is the exported format?

  • @nazilaameli8639
    @nazilaameli8639 Год назад

    Thanks for the fantastic presentation! the prediction gives me black images, how can I fix it? I am working on tooth radiographs trying to segment the crowns. thanks

  • @banjerdsa-ngiem1510
    @banjerdsa-ngiem1510 2 года назад

    Can I train with image 1080*1080 ?

  • @nataliameira2283
    @nataliameira2283 Год назад

    What tool did you use to label the data?

  • @subhayanmondal735
    @subhayanmondal735 Год назад

    Thank you so much Sir for such a great presentation... Sir, can I do segmentation of any image size with this code? I am working in the field of radiation oncology and I am working on organ segmentation from CT/MRI images. I need your valuable suggestions regarding the same.

  • @matancadeporco
    @matancadeporco 3 года назад +2

    i'm did this with my datased containg images of tomato leaves, got IOU score of 1.0 (thats bad right?) and on the prediciton plot, i got nothing, just a black square... what u think its going wrong? ty 4 ur videos, its helping me a lot

    • @ZEISS_arivis
      @ZEISS_arivis  3 года назад +1

      IOU of 1.0 means you are getting 100% overlap between prediction and labels. This is unusual! Please make sure your are indeed comparing predictions and original masks.

    • @adnanebadaoui4033
      @adnanebadaoui4033 2 года назад

      me too i have black square help please

  • @불루이보스
    @불루이보스 2 года назад

    Thanks a lot! Your really good teacher, sir!!

  • @nibhanapuri4476
    @nibhanapuri4476 3 года назад

    Code used in video is not available in your github repo and please share

    • @ZEISS_arivis
      @ZEISS_arivis  3 года назад

      Just checked, it is there. Please check again.

  • @fuegopuro5933
    @fuegopuro5933 3 года назад

    Can you go with BRATS pls?

  • @shabinaa6407
    @shabinaa6407 3 года назад +1

    Sir, it works very well on Corpus callosum segmentation, and for Lateral Ventricle segmentation it gives only black image and shows error: predictions` out of bound
    Condition x < y did not hold, in IOU_keras.update_state(y_pred_thresholded, y_test).

    • @naiyraelkady8204
      @naiyraelkady8204 3 года назад

      can u tell me if this is considered a pre-trained model or not

    • @shabinaa6407
      @shabinaa6407 3 года назад

      @@naiyraelkady8204
      No, pre-trained means model is trained and saved, you have to just call that model and give your input. before providing input u must match your input size.

    • @shabinaa6407
      @shabinaa6407 3 года назад

      I got segmentation after changing a number of filters. thank you @Apeer_micro

    • @ZEISS_arivis
      @ZEISS_arivis  3 года назад

      I am glad you figured out the problem and fixed it.

    • @antoniojunior4853
      @antoniojunior4853 3 года назад

      I'm having the same problem (getting black image prediction)... and I cant figure that out.. I tried to change the number of filters but it didnt solve my problem..

  • @raiden4743
    @raiden4743 2 года назад

    I am getting an error "ValueError: Found input variables with inconsistent numbers of samples: [80, 0]" can anyone help me with this. I am new to this field.

  • @jocelynw6360
    @jocelynw6360 3 года назад

    Sir i am new to this, so I am trying to segment CXR using Unet. I found some data from kaggle but it doesn't contain the mask part.. How do I extract the mask image sir.. I really need help sir..

    • @aravindramkumarkrishnan8135
      @aravindramkumarkrishnan8135 3 года назад

      Hey there! There is a video by a person named Seth Adams. He explains how masks can be created from images. Here's the link to the video:
      ruclips.net/video/udR6SwojYXo/видео.html
      I hope you find this useful!

    • @yordanosbirhanu483
      @yordanosbirhanu483 2 года назад

      u have to segmnetation the image first do annotation

    • @EmilTheLongshoreman
      @EmilTheLongshoreman Год назад

      Use something like label studio to annotate the CXR images with 'areas of interest' then export the mask for training, there are a bunch of tutorials how to do this online. not ideal because you have to manually do this step sometimes but this is one way to accomplish your task here

  • @bourdayrachid3855
    @bourdayrachid3855 2 года назад

    thank you teacher for your interssent explication can you please share with me the sildes for architecture unet Used

  • @dantec.dagandanan3732
    @dantec.dagandanan3732 3 года назад

    thank you so much ser.

  • @siobhanlight7109
    @siobhanlight7109 2 года назад

    Thank you so much!

  • @sergiotalegon9868
    @sergiotalegon9868 2 года назад

    THANK YOU :)

  • @gampangji6171
    @gampangji6171 2 года назад

    Pardon me, Can I get you Email Please ?? thank you so much