Mastodawn

Today I tested Segment Anything Model (RX 6700 XT / 16 GB RAM)

SAM was developed by Meta and is a local FOSS AI model for image segmentation. It can detect objects in images and generate precise masks, making it possible to isolate people or objects from the background.

In my test, I automatically selected the largest detected mask, isolated it, and processed it further. This allows subjects to be separated quickly from the background for image editing, compositing, stickers, memes, or creative video effects.

Model used: sam_vit_h_4b8939.pth

SAM can be used for:
- Object cutouts
- Image editing & photomontage
- Video compositing & post-production

A newer version, Segment Anything Model 2, extends these capabilities and is especially designed for video segmentation, providing better temporal consistency across frames and more stable object tracking over time.

Video workflow:
- Recorded with OBS
- Edited in Kdenlive
- Transcoded with VAAPI (H.264)

No cloud, real hardware.
Everything runs on Linux, so anyone can set this up.
Works on CPU as well, but much slower.

Background music: Sweet but Psycho - Ava Max [Rock Version by Kenke] (https://www.youtube.com/watch?v=714BjMMUi-s)

#AI #MachineLearning #ComputerVision #ImageSegmentation #SAM #SegmentAnything #MetaAI #FOSS #OpenSource #DeepLearning #ImageEditing #Compositing #VideoEditing #Tech #Innovation

hansakwast Mar 19

Review of the QGIS GeoAI plugin

https://videos.qwast-gis.com/w/7AErmuFVuMvnG4wGk3EYtJ

Review of the QGIS GeoAI plugin

PeerTube

feitgemel Mar 13

Build custom segmentation models with YOLOv8 and SAM! 🚀

Full guide: https://eranfeit.net/segment-anything-tutorial-generate-yolov8-masks-fast/

Watch: https://youtu.be/8cir9HkenEY

#Ultralytics #SegmentAnything #AI

sayzard Feb 14

無異 (@taekie)

메타의 SAM(Segment Anything Model)을 시험해 본 결과를 공유. 스포츠 중계 영상에서는 카메라 시점과 줌 변동을 추적해 스트레칭 처리하는 기술이 핵심이라고 설명하며, 관련 예시로 작년 로잔 박윤재 발레리노 영상을 링크함.

https://x.com/taekie/status/2022565770543403222

#meta #segmentanything #computervision #tracking

無異 (@taekie) on X

메타 SAM 잘되는지 한 번 만들어 봄. 스포츠 중계 영상에서는 카메라 시점이나 줌이 바뀌어서 그걸 추적해서 스트레칭해주는 것이 핵심 기술인 듯. 영상은 작년 로잔 박윤재 발레리노. https://t.co/S1eNNSc5xz

X (formerly Twitter)

hansakwast Jan 29

Review of the QGIS GeoAI plugin

https://videos.qwast-gis.com/w/dMS9CD17L8WEfnmi5AWEJ2

Review of the QGIS GeoAI plugin

PeerTube

TechGlimmer Nov 21

Meta’s SAM 3 is the new AI superpower for visuals. Find & select anything in photos/videos, just by typing, clicking, or dropping an example. No more tedious edits, just instant results.

Check out my fun explainer here:
https://techglimmer.io/meta-segment-anything-model-3-sam3-explained/

#Meta #AI #SegmentAnything #Sam3 #PhotoEditing

Meta Segment Anything Model 3: The Easy Way to Find Anything in Your Photos and Videos

Meta Segment Anything Model 3 (SAM 3) finds any object in photos and videos using text or clicks. Free access, easy to use. Learn how it works

techglimmer.io

AI Daily Post Nov 21

Meta’s latest SAM 3 model shows promise but stumbles on niche technical terms and complex logical prompts. While its zero‑shot abilities shine on general images, medical‑imaging tasks and 3‑D segmentation still lag behind Llama and Gemini. Find out what this means for open‑source vision research and where the community can help improve it. #MetaSAM3 #SegmentAnything #ZeroShotAI #MedicalImaging

🔗 https://aidailypost.com/news/metas-sam-3-falters-niche-technical-terms-complex-logical-prompts

Reddit Tech VN Bot Nov 19

Meta ra mắt SAM 3: Công nghệ phân đoạn hình ảnh với concept.
#SAM3 #Meta #TríTuệNhânTạo #AI #SegmentAnything
#CôngNghệMới #TríTuệArtificial #MetaSuperintelligenceLabs

https://www.reddit.com/r/LocalLLaMA/comments/1p1df5y/sam_3_segment_anything_with_concepts_by_meta/

Stefan Ihringer Apr 28, 2025

I'd love to work on shows where SAM2 is good enough. But we have actors with fine hair detail and the comps need to survive 4K inspection so.... Indian roto vendors it is :-/

https://floss.social/@kdenlive/114416926889657357

There's a #SegmentAnything toolset for #Nuke: https://github.com/Theo-SAMINADIN-td/NukeSamurai

It works but the lack of temporal stability makes it break down when actors are not moving. All the cool test footage is people running or objects moving around which is much more forgiving.

Kdenlive (@[email protected])

Attached: 1 image Kdenlive 25.04 is out with background removal (SAM2), OpenTimelineIO import/export, performance enhancements, optimized audio waveforms and lots workflow improvements and bug fixes. https://kdenlive.org/news/releases/25.04.0/ #kdenlive #floss #videoediting #artWithOpenSource #OTIO #pixar @[email protected]

FLOSS.social

Harald Klinke Jul 30, 2024

#segmentanything #Cloisonismus