Auto Annotation for generating segmentation dataset using YOLOv8 & SAM

Learn how to auto annotate the Image Segmentation dataset.

#####################################################
####################################################

In last Month (April 2023), Meta AI released the Segment Anything Model (SAM) which is an instance segmentation model. Segment Anything was trained on over 1 billion masks on 11 million images. This is the largest segmentation dataset to date. SAM model can possibly segment any objects.

Recently Ultralytics added this SAM model inside their Ultralytics package and named it Auto annotation. Auto-annotation is a feature that allows you to generate a segmentation dataset using a pre-trained detection model. We can easily annotate large datasets with it. And there would be no need of manual annotation. It will save lot of time and effort.

#objectdetection #imagesegmentation #sam #yolov8 #computervision

[Sunil-ezhx]

What a great video ma’am. Very good presentation skills. Thanks

[nithinkumar-cs]

Great work! Thanks for teaching for free

[nouhailamokhtatif]

Hello, thank you for the vide, i want to know the nature of the output generated by sam, the first index is the object class and what about the other numbers?

[olanrewajuatanda]

Please, how do I serve multiple images and or image sequence as input and how do I save the output?

[kthakrar]

How do you think this performs relative to using them independently?

[karlm]

This is great and thank you so much! Note that the detection model used must be a YOLOv8 detection model. I attempted to use a YOLOv5 model and it would not work. I am still training the YOLOv8 model on my dataset (5 days!), and after it is completed I will attempt to use it to perform the auto annotation. Thankfully the annotation format for YOLOv8 and YOLOv5 is the same for my 23, 000 images, but the data.yaml is slightly different.
(I don't want to annotate 23, 000 images for segmentation manually, it already took me 3 months to annotate them just with bounding boxes!)

[soravsingla]

Hi Aarohi, your content is excellent and your channel is one of the best Artificial Intelligence channel but still not getting that much of likes which your channel deserves. Hope you succeed #AI
#ArtificialIntelligence
#DataScience
#EducationalContent

[nguyenviethoan]

thank you for your tutorial, it is very nice. Can I ask a question ? The segmentation labels are saved as txt. If I want to edit the auto-annotation segmentation, which labeltool can recognize these annotated txt files ?

[raj_]

amazing, please also make some tutorial video on 3D point segmentation.

[bouchakourlallouani]

How to use the yolo for texte detection

[zuwrjom]

How do you run the model in Colab with a camera?

[asfandiyar]

Thank you for video. The git repo has become dated unfortunately and I had to do a little bit of work to get this working. I'll try to merge my work later. Thanks again.

[jeffreyeiyike]

Good day, I would like to extract yolo feature. What is the best way to do it

[ashirahmed]

Hi, how can i extract these segments from the image and make new images with just the segment

[ZeeshanAli-cltb]

I am using yolo for detecting the fire
I have dataset of 10000 images how could I annotate all of the images

[rangurangu]

Kognic annotation 2d images how to use this code auto select segmentation on running kognic web task...can you please tell me.

[reebalnofal]

Thanks a lot for your videos, i was wondering how can i use SAM if i already have the bounding boxes labels as [class x1 y1 x2 y2], should i use the last part of your code in this case?

[ZhCrypto]

What does mean by image segmentation and where is code

[SVajrapuPrathyusha]

Hi mam... I am following your channel.
Your videos are amazing🤩
I watched your traffic sign detection using yolo v4 and doing the project on that...
I have some doubts in the code..
I posted the comments in that video. Please give your answer to my question which i posted their
Thank you

[rama_bhuyan]

Hi
Thank you for the great explanation.
I got a question- how do we get the bounding boxes for costume dataset ( class not in COCO/IMAGENet). Does the yolo model has to be trained/ if yes then how many minimum manual annotations required so that we could use this tool effectively/efficiently.