Amazing tutorial. Thank you. You are using test data to validate and to test. Should we not use some part of train data to validate and use test data only for testing?
Thank you for the great tutorial. I downloaded the dataset but the test folder only contains images and mask folders not the 1st- and 2nd-manual folder. Do you know by any chance where can these be downloaded?
Why would you augment images and masks separately and save them? You can use augmentation techniques in dataset class so that at every iteration the model sees different images. Iterating for more epochs, you can give the model more augmented images so that the model does not overfit.
What if I don’t have the images that you have in the 1st manual of your dataset. Is there any way to create them?. Actually i am using Messidor Dataset. Can you help? I really need to understand
It's really helpful vedio. I am learning through this video and trying every bit of it. But I am getting the length of training and testing as 0. Whereas you ate getting it 20 which is the actual size.
I got the mistake. On your first script (data_aug), change the "test_y = sorted(glob(os.path.join(path, "test", "1st_manual", "*.gif"))) for "test_y = sorted(glob(os.path.join(path, "test", "mask", "*.gif")))" It may work
Nice video , I used your codes for radiography segmentation but the loss decreases very slow and it doesn't reach the optimum amount , do u have any advise ?
I used same code for segmentation of microanerism,hard exudates and haemorages,l got decent outputs but i am getting jaccard,F1,Recall, precision as 0 and i am getting good accuracy when using threshold value 0.5 but other metrics are zero, what might be the problem?. please help
@@jaiganeshshetty4883 brother try to look for a good dataset , if you are using his dataset then i am not sure what's the problem but i usually get problems in dataset and its path(between images and masks) and when i am using his augment_data function i am also getting a error of cv2.error: ... src data type = bool is not supported , my augmented images are not storing do you know how to resolve this ?
Anyone who tried to execute this in Collab or Kaggle please respond,even there also I am getting the same error..I need this for my final year project ,if any please help
if you are on Windows, the "\" or "/" for the file paths are different, therefore the os.path.join function and glob function insert the wrong slash. To avoid the problem, use the following code for the load_data function: def load_data(path): train_x = sorted(list(map(lambda x: x.replace('\\', '/'), glob(os.path.join(path, "training", "images", "*.tif"))))) train_y = sorted(list(map(lambda x: x.replace('\\', '/'), glob(os.path.join(path, "training", "1st_manual", "*.gif"))))) test_x = sorted(list(map(lambda x: x.replace('\\', '/'), glob(os.path.join(path, "test", "images", "*.tif"))))) test_y = sorted(list(map(lambda x: x.replace('\\', '/'), glob(os.path.join(path, "test", "1st_manual", "*.gif"))))) return (train_x, train_y), (test_x, test_y)
Such a big help for my final term project! Thank you!
Nice video covering every single detail. Thanks
Amazing tutorial. Thank you. You are using test data to validate and to test. Should we not use some part of train data to validate and use test data only for testing?
Really helpful..thank you😊
Respect and amazing work bro,,, keeping doing it!!
Appreciate it!!
Can you do a video on retina lesion segmentation using deep learning
Thank you for the great tutorial. I downloaded the dataset but the test folder only contains images and mask folders not the 1st- and 2nd-manual folder. Do you know by any chance where can these be downloaded?
why do we have two loss functions ?
Why would you augment images and masks separately and save them? You can use augmentation techniques in dataset class so that at every iteration the model sees different images. Iterating for more epochs, you can give the model more augmented images so that the model does not overfit.
Thanks for your suggestions. It was an old video. Now, I use real-time augmentation.
great work
Thanks. I am glad you liked it.
What if I don’t have the images that you have in the 1st manual of your dataset. Is there any way to create them?. Actually i am using Messidor Dataset. Can you help? I really need to understand
it's so helpful for me. Thank you. I have a question about train.txt file, how you generate it. How we can work with checkpoint.pth properly?
How can we add training and validation loss graph using the checkpoint file?
It's really helpful vedio.
I am learning through this video and trying every bit of it. But I am getting the length of training and testing as 0. Whereas you ate getting it 20 which is the actual size.
Same thing with me
I got the mistake. On your first script (data_aug), change the "test_y = sorted(glob(os.path.join(path, "test", "1st_manual", "*.gif"))) for "test_y = sorted(glob(os.path.join(path, "test", "mask", "*.gif")))"
It may work
For data_aug.py file my train runs and it gives 100% in the IDE but when I see folder there are no image that has been saved. Can you help me out here
I think you are using Windows. In Windows, there is always some issues with the path. Check the path.
Hey, I have the same issue. How did you solved it?
i have the same issue
are the mask pixels range in 0-1 or 0-255?
Mask pixels are in the range of 0-1
Nice video , I used your codes for radiography segmentation but the loss decreases very slow and it doesn't reach the optimum amount , do u have any advise ?
Plz Make a video on RAVIR dataset for Retinal Image Segmentation.❤
If it is possible can you please add the link to the dataset on the description.
I used same code for segmentation of microanerism,hard exudates and haemorages,l got decent outputs but i am getting jaccard,F1,Recall, precision as 0 and i am getting good accuracy when using threshold value 0.5 but other metrics are zero, what might be the problem?. please help
If I use the threshold as 0.1 i get the values but they are low
My torch version:2.5.1+cu124
Can this cause problem?
@@jaiganeshshetty4883 brother try to look for a good dataset , if you are using his dataset then i am not sure what's the problem but i usually get problems in dataset and its path(between images and masks) and when i am using his augment_data function i am also getting a error of cv2.error: ... src data type = bool is not supported , my augmented images are not storing do you know how to resolve this ?
sir i don't " cuda " it is giving me error
hey I am implementing same code on kaggle .but It is giving some errors can you help me to rectify
from model import build_unet, from loss import DiceLoss, DiceBCLoss, and from utils import seeding, create_dir, epoch_time did not run. Kindly help
Hi,I am getting CUDA out of memory error, please help me to solve this problem.. please
Anyone who tried to execute this in Collab or Kaggle please respond,even there also I am getting the same error..I need this for my final year project ,if any please help
Try to reduce batch size
I ran the same code, yet it is not saving the images to the new data folder. Can you please help?
I think you are using Windows. Please check your path. The slash are different in Ubuntu and Windows.
Did you find a solution for this ?
if you are on Windows, the "\" or "/" for the file paths are different, therefore the os.path.join function and glob function insert the wrong slash.
To avoid the problem, use the following code for the load_data function:
def load_data(path):
train_x = sorted(list(map(lambda x: x.replace('\\', '/'), glob(os.path.join(path, "training", "images", "*.tif")))))
train_y = sorted(list(map(lambda x: x.replace('\\', '/'), glob(os.path.join(path, "training", "1st_manual", "*.gif")))))
test_x = sorted(list(map(lambda x: x.replace('\\', '/'), glob(os.path.join(path, "test", "images", "*.tif")))))
test_y = sorted(list(map(lambda x: x.replace('\\', '/'), glob(os.path.join(path, "test", "1st_manual", "*.gif")))))
return (train_x, train_y), (test_x, test_y)
@@anagha2765 hi did u get the solution?
bro you saved me with this code
@@arthureglin
Hi could you also do a video on 3D unet
Which tool did u use to annotate the images?
The data is automatically annotated.
labelstudio
where can i get the data?
Which Notebook you are use?
did u find it?
@@taharabs8006 atom idle
atom idle
Can you do image segmentation on nii.gz files
I haven't started working on 3d segmentation yet.
mask = mask / 255.0
TypeError: unsupported operand type(s) for /: 'NoneType' and 'float' please, tell how to solve this error
Hello, where can I get the dataset
Mate if u found it please let me know !
GOT THİS ERROR
TypeError: unsupported operand type(s) for /: 'NoneType' and 'float'
Please check the path of images and masks. The cv2.imread is not able to read the image.
What did you use to annotate the retina images so intricately?
thresholding and Cleaning,
Your email
nikhilroxtomar@gmail.com
If it is possible can you please add the link to the dataset on the description.
mate did u find it ? if u did please let me know as well please