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
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...
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
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
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
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.
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
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.
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 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.
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..
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.
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..
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!
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
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
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...
did u guys figure it out !!
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 ?
Sir,what should i do to get the confidence of the object?
Thx you very much
Dear Sreeni sir , this video helped me so much .. thanks a lot
Thx you very much for your example and your nice explanation. Hope to see more content from you!
Thanks, will do!
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
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
What is the exported format?
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
Can I train with image 1080*1080 ?
What tool did you use to label the data?
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.
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
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.
me too i have black square help please
Thanks a lot! Your really good teacher, sir!!
Code used in video is not available in your github repo and please share
Just checked, it is there. Please check again.
Can you go with BRATS pls?
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).
can u tell me if this is considered a pre-trained model or not
@@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.
I got segmentation after changing a number of filters. thank you @Apeer_micro
I am glad you figured out the problem and fixed it.
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..
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.
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..
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!
u have to segmnetation the image first do annotation
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
thank you teacher for your interssent explication can you please share with me the sildes for architecture unet Used
thank you so much ser.
Thank you so much!
THANK YOU :)
Pardon me, Can I get you Email Please ?? thank you so much