Segment Anything - Model explanation with code
ฝัง
- เผยแพร่เมื่อ 28 มิ.ย. 2024
- Full explanation of the Segment Anything Model from Meta, along with its code.
As always the slides are freely available: github.com/hkproj/segment-any...
Chapters
00:00 - Introduction
01:20 - Image Segmentation
03:28 - Segment Anything
06:58 - Task
08:20 - Model (Overview)
09:51 - Image Encoder
10:07 - Vision Transformer
12:30 - Masked Autoencoder Vision Transformer
15:32 - Prompt Encoder
21:15 - Positional Encodings
24:52 - Mask Decoder
35:43 - Intersection Over Union
37:08 - Loss Functions
39:10 - Data Engine and Dataset
41:35 - Non Maximal Suppression - วิทยาศาสตร์และเทคโนโลยี
As always the slides are freely available: github.com/hkproj/segment-anything-slides
I am a undergraduate student learning AI, your videos help me a lot. Thanks for the selfless work you are doing.
Please keep making this kind of videos Umar! You're really a gifted teacher.
Thank you for your kind words, @pierret00
Please stay tuned, more videos coming soon
I am a undergraduate student learning ML, your videos help me a lot. Thanks for the selfless work you are doing.
Amazing explanation!
23:39 -> loved that intuitive explanation!
It's amazing how you can make this difficult subject much easier and more fun to learn! Thank you!
Very helpful, thanks! Please keep uploading 🙂
Great explanation man!
Thank you! This is really helpful
Really great explanation !! Thanks !
awesome video man!
Hi Umar, thanks for this amazing code explanation. Just one question, how is the prediction_iou computed in the Automatic Mask generation of SAM? I am asking because we only have the model's prediction and to compute iou you need ground truth labels. Thanks!
Thank you!!!
This is very helpful
The video is awesome.
incredible explanation! would you be interested in reviewing the SEEM (Segment Everything Everywhere All at Once) model?
A video for coding SAM from scratch would be entertaining!
One question regarding the Vision Transfomrer, even if this is not the main topic:
Why do they use the FIRST Token? Wouldn't usually the LAST token caputre all the information of the before seen ones?
thanks a lot. how can we access to codes that explained in video?
Nice video!
I think your understanding of the source code is really good.
Could you please share your annotated code? It will be very helpful for my undergraduate graduation project.
Is there any way we can mask same object with same colors ?
Valeu!
based on python roadmap, what topic should i focused to sir?