2. About me
❏ Yurii Pashchenko
❏ Principal Machine Learning Engineer at Depositphotos
❏ Over 10 years of research and commercial experience in
applying Deep Learning models
❏ Object Detection/Segmentation and Face Recognition
Specialist
3. Unlocking the potential of Segment Anything Model
● Image segmentation
● Segment Anything Model
● Examples of use
● Limitations
5. Basic Image Segmentation
“Image segmentation is a sub-domain of computer vision and digital image processing which aims at
grouping similar regions or segments of an image under their respective class labels”
https://www.v7labs.com/blog/image-segmentation-guide#h3
10. Segment Anything Model (SAM)
SAM: A generalized
approach to
segmentation
https://ai.meta.com/blog/segment-anything-foundation-model-image-segmentation/
12. SAM.Task
Prompt -> Valid mask
prompt can be
● a set of foreground/ background points
● rough box or mask
● free-form text, or, in general, any information
indicating what to segment in an image.
“valid” mask simply means that even when a
prompt is ambiguous and could refer to multiple
objects the output should be a reasonable mask for
at least one of those objects.
14. SAM.Dataset
SA-1B consists of 11M diverse, high-resolution, privacy protecting images and 1.1B high-quality segmentation
masks that were collected with our data engine.
27. Meta didn't release the text prompt feature for SAM
Text Encoder.CLIP
CLIP Surgery for Better Explainability with Enhancement in Open-Vocabulary Tasks
28. Meta didn't release the text prompt feature for SAM
Text Encoder.CLIP
https://github.com/xmed-lab/CLIP_Surgery/blob/master/demo.ipynb
30. Semantic and Panoptic Segmentation
A pipeline for panoptic segmentation can be like this:
1. Use Grounding DINO to detect the "thing" categories (categories with
instances)
2. Get instance segmentation masks for the detected boxes using SAM
3. Use CLIPSeg to obtain rough segmentation masks of the "stuff" categories
4. Sample points in these rough segmentation masks and feed these to SAM
to get fine segmentation masks
5. Combine the background "stuff" masks with the foreground "thing" masks
to obtain a panoptic segmentation label
https://github.com/segments-ai/panoptic-segment-anything
35. Thank you for your attention!
AI&BigData Online Day 2023
Yurii Pashchenko
Principal Machine Learning
Engineer at Depositphotos
yurii_pas
george.pashchenko@gmail.com