Labeling images for semantic segmentation using Label Studio
HTML-код
- Опубликовано: 19 июн 2024
- The code snippet for this video can be downloaded from:
github.com/bnsreenu/python_fo...
For other code available on my GitHub:
github.com/bnsreenu/python_fo...
For labeling your images using Label Studio:
labelstud.io/
Let us work in Anaconda command prompt. (You can use other command prompts)
Check environments:
conda env list
Create a new environment to install Label Studio:
conda create --name give_some_name pip
(Need to specify pip as a dependency, otherwise it will not be available)
(To specify python version for your env..)
conda create -n give_some_name python=3.7
Now activate the env.
conda activate give_some_name
Install the Label Studio package
pip install -U label-studio
Launch it!
label-studio
Open your browser and go to the URL displayed on your screen, typically
0.0.0.0:8080/ 
Наука
I wish there were more people out there like you. You explain everything very well, and you aren't rushed. Great work
yes, well said
Brilliant as ever! I have learnt so much about computer vision from this one channel and it has helped me in both academic and professional contexts. I can’t thank you enough for dedicating your time to producing such high quality content.
Thank you for all the great videos!
I checked a bunch of videos and articles on building a custom dataset for a segmentation problem. Your video is by far the clearest, informative, and the most concise for a beginner like me. Other resources I had a bunch of questions after going through, but yours, zero question so far. I will try it out now. Thank you so much!
Awesome, thank you!
Did not expect this tutorial from you, Aj.Sreeni.......Still watching and once it is done, will post by comments.....Thanks a ton for this.....
thank u,great job,the way u explained every thing is very simple and neat.u are the great motivator for me to do the coding
I did the same work with Paint 3D which are build in Windows :D and saved the result as Tiff image. They will not have bluring edge. But label studio is a more professional tool,classes can be modified and added more easily, Hopefully this will be improved in the future, thank you for sharing this tool, it is very useful indeed. cheers
Your videos are an awesome contribution! Keep it going!
Glad you like them!
Thank you very much. This video was very useful!
Great! This is a useful guide for learning, Thank you very much!
Glad it was helpful!
Sir, you are doing great service to mankind by educating them
It's my pleasure
Thank you for introducing us to a useful and new tool.
Which of the existing tools or the tools you introduced this time should be used first?
Thanks for the great content. Is it possible to further post-process these images and determine the properties of labelled data? Such as the curvature of the road, the orientation of the road from the horizontal axis, width, height quantitatively.
Amazing work Sreeni. Kindly show us how to segment colour images trained with patched U-Net.
Realy Thanks for pretty helpful vieo
Thank you for the awesome tutorial.
Can create a classifier model by label-studio so you can apply it to more images?
Thank you very much!
very helpful as always, thank you
My pleasure!
Good, it is very helpfull.
Many thanks for this !
You are welcome!
Great job
Thank you for making such a wonderful informative videos. Which tool support labelling images with tiff format having 4 bands?
Thank you so much this gonna help me a lot
Glad to hear it!
Thank you fir the video, so we need to install sqllite before?
Great Video :)
Thank you for the awesome tutorial.
I want to use a national land cover images as mask (thematic single-band .tif) with 6 labels, how can I convert it to useable image for for semantic segmentation???
Hello. I watched all your videos and they are great! Thanks for all your support on labelling. But I need something else and I couldn't find any information about my task.
I have an satellite image and I need to extract all the road marks from the road.
But I want to do it faster and automatically. I tried with CVAT, but their software doesn't help me with that.
Can you suggest me something open-source and user friendly?
Thanks alot!!
How to deal with the jagged edge in output (during segmentation). I am providing perfect mask for segmentation but still getting jagged edges from my trained model. I am having this issue from long time. Please suggest something. Thank you.
The masked images that I exported have different file names from my original images. Is there an option to rename the mask files so that I can map them to the original images? Thanks.
Did you managed to this situation?, I have so many images and need to obtain their masks with their names.
Okay so after we annotated can we train a model to detect the annotated region using mask rcnn or yolo v7?
How much does the annotation at edges of class using brush effect in the accuracy of the segmentation model ? Labeling the corner edges of monuments right is troubling me.
Thank you!
Thank you very much.
For multi-class segmentation, for example 3 as per your video, are we supposed to have 3 different PNG images for each class for that one particular image?
I’m new to this and what I’ve done was give each label a different colour, (blue, red and yellow, and background as black) and exported the mask as a PNG file, in which the resulting mask is just one PNG file with a black background and the 3 colours. Is that a wrong way to do it?
one image for satellite imagery, but multiple mask images, so each class has a mask. Then you would do multiple satellite images, each with masks for multiple classes.
after labeling, we got two things brush label for image and NumPy, what are the next steps? how to train the data so that similar images can be detected?
Hello Sreeni, Do you how to Label a CBCT Scan for 3D segmentation?
Just a question, as you are working with geospatial data why would you not use qgis for example and create shape files? Is it just more straightforward like this? Great channel btw
I have the same question. Besides, is the geo-location still remined in the labled image?
same question, are these labels still georeferenced?
I want to use the Json format. Do I need to do changes similar to the binary?. Thanks in advance!!
Hi Sir.. Can you make videos on how to recreate/modify a pre-annotated data in label studio?
Thanks
Hi Sreeni how are you doing? I'm rewatching your videos about annotation, I about to start a big project of image segmentation where I will need to annotate like more than a thousand images. What solution are you using currently and what you suggest in this case? I have label studio installed, but I feel it is kind not that practical. It is breast cancer project where I will need to segment the breast into several portions or quadrants.
Why not practical?
Hi Sir, I had a very important question regarding labelling. I have an image with hair to label for annotation, if one image has 30 hair and them too overlapping but the hair are all very similar when it comes to colour and thickness. If i label 3 or 4 of these hair and go onto another image to annotate with a different sort of hair. The main question is if I label just few object of interest even though there are more of them in the image. Is it still good annotation ? because I have a huge dataset which I need to annotate and I dont have much time. So I am literally labelling just 4 or 5 hair from each image even though there are more hair in the image. Is it okay or not ??
please reply to this!!
Thanks. Is it possible to call a python function during the labeling process for auto-labeling? I have a code that learns a classifier using the already labeled pixels and do the rest of the labeling itself. Then I correct the wrong label predictions & update the classifier. This way it can be done interactively. Moreover, if you have thousands of images, you can run the auto labeling code on all of them & update the correct labels iteratively.
is this active learning?
@@garfield111garfield Yes
How the YOLO format will be in label studio could you please tell me.
What if the images had edges which had different colours like red and black or blue and balck in the same image how will we normalize the edges then?
Beginner here, Is there link to explain how to train using the above annotated images?
the corrresponing file names have difeerent names from the images after exportation ! will it create problem?
I wanted to measure an angle between two bones of an x ray image, how to measure angles?
I'm planning to do coffee grading research using deep learning approach, but I'm stacked in the middle of selecting the proper technique used for the task.
The actual coffee bean grading is performed based on defect counts found in 350gm bean sample. There are multiple grades of coffee beans(country specific), in each grade there might be different types of defects with more than one occurrence. In this case how could we prepare a datasets and what is the proper technique used for image processing?
I would be grateful if you help me.
Are you trying to find defects from images or using already existing defect information for coffee grading? If you are trying to find defects in images, then I have a few videos on anomaly detection - ruclips.net/p/PLZsOBAyNTZwYE5IM1g_3yNXuxaH_QRSoJ
@@DigitalSreeni using existing defect information
@@DigitalSreeni is it possible to do both tasks; defect detection and classification of grade based on defect counts?
Please make few more video in label studio .
thanks bruh
No problem
Is it better to label each label as 0 and 1 or to have multi labels on one image where houses is 1, water 2, etc? Would that limit you to 255 items?
I guess you are asking why the host used np.where to set 1, 2, and 3 in this video. In my view, the value that the host set to from 1 to 3 was not the labels but the intensity values of the image by leveraging numpy.where filter method. It is a kind of binary threshold that converts a range of values to a specific value if they were satisfied the condition. I guess that the host wanted to feed those different classes into the nn using different intensity value for easy learning.
hii..how do u make sure the export labels match the imported files..am stuck here.
How do I label images if I have count crowd ?
Export to PNG option is disabled on my LableStudio... please let me know how to enable it? thanks!!
is it possible to upload tiff files and annotate?
Is it possible to use label studio to make instance segmentation labels?
Yes, I believe so.
any chance you can use photoshop? then import it.
thanks, but i have problem when i install it , like this note: This error originates from a subprocess, and is likely not a problem with pip.
error: metadata-generation-failed
Hello I would to import my labeled mask to correct in Label Studio. How can I do it ? Thank for our help
Looks like this is a question for Label Studio support. I don't know the answer.
Can you please make a video on Graph Neural Network for brain tumor images segmentation. Thanks in advance.
hi sir, can sematic segmentation be used to segment pedestrians from background
Yes, but I recommend instance segmentation as you are trying to detect individual objects. Try Mark R-CNN.
@@DigitalSreeni okay sir thankyou for your suggestion.
could you please share the code here in the comments box?
Thank you!
Thank you very much for being generous.