Meta releases its Segment Anything Model and Segment Anything 1-Billion mask dataset, hoping to help researchers with computer vision and object identification
and Meta is sharing the code Katie Paul / Reuters : Meta releases AI model that can identify items within images GitHub : Segment Anything — Meta AI Research, FAIR — [Paper] [Project] [Demo] [Dataset] [Blog] Alexander Kirillov / Meta AI : Segment Anything — Abstract — We introduce the Segment Anything (SA) project: a new task … Sakshi Goyal / Analytics Insight : Facebook-Owner Meta Releases AI Model to Detect Items Within Images Livemint : Meta unveils AI model for object detection and largest dataset of its kind Saundra Latham / LinkedIn News : Meta unveils image-analyzing AI Global Village Space : Meta unveils AI model for detecting unseen objects Alkesh Sharma / The National : Facebook owner Meta introduces AI model that can identify objects in images There's An AI For That : Segment Anything by Meta Lawrence Bonk / Lifewire : Meta's Nifty New AI Photo Segmentation Tool Cuts Anything Out of an Image LinkedIn: Kiryl Trembovolski : This is magical. Meta has just released Segment Anything Model (SAM), the first foundation model for image segmentation. … Tweets: @metaai : Today we're releasing the Segment Anything Model (SAM) — a step toward the first foundation model for image segmentation. SAM is capable of one-click segmentation of any object from any photo or video + zero-shot transfer to other segmentation tasks ➡️ https://ai.facebook.com/... https://twitter.com/... @drjimfan : Reading @MetaAI 's Segment-Anything, and I believe today is one of the “GPT-3 moments” in computer vision. It has learned the *general* concept of what an “object” is, even for unknown objects, unfamiliar scenes (e.g. underwater & cell microscopy), and ambiguous cases. I still can't believe both the model and data (11M images, 1B masks) are OPEN-sourced. Wow.😮... Anshel Sag / @anshelsag : Maybe people will realize that a lot of @Meta's R&D also comes from it's extensive #AI research which both benefits it's social media apps as well as XR ambitions. Meta is not a follower in the AI space, it is a leader. https://twitter.com/... Johannes Otterbach / @jsotterbach : Foundation Models (FM) for language, aka #LLMs, taught us a lot about the completeness and power of language abstractions. I wonder what FM for vision will teach us? The future is really amazing 😎 https://twitter.com/... Markus Frey / @cyhsm : @ylecun This seems to solve my recently proposed benchmark which *should* have been hard for segmentation models: https://twitter.com/... Not much room for competition when you're a compute-constrained researcher contesting foundation models 💻🤷♂️ https://twitter.com/... Danfei Xu / @danfei_xu : One of the most impressive CV works I've seen recently. Also huge kudos to Meta AI for sticking to open sourcing despite the trend increasingly going towards the opposite direction. https://twitter.com/... @iandanforth : I'm impressed! It does a fantastic job on what I thought would be a hard task, book segmentation on a casual image! https://twitter.com/... https://twitter.com/... Eric Jang / @ericjang11 : Thank you @MetaAI for open-sourcing the model + code under Apache 2.0 👏👏👏 More like this pls! https://github.com/... https://twitter.com/... Terry Xu / @coolnalu : Love this: integrating this with the stable diffusion pipeline and generate images in parallel could mean even more astonishing image and video generation freedom https://twitter.com/... Alexandre Cadrin-Chênevert / @alexandrecadrin : Impressive but imperfect zero-shot generalization in medical imaging for Segment Anything Model (SAM) from FAIR https://segment-anything.com/ https://twitter.com/... https://twitter.com/... @dazzkraz : Mega models will eat up start ups. Until all these years we had these start ups going after industry use cases focusing AI on niches. Now you have these giga models which out perform them all , it's interesting to see how the economy settles in the coming months https://twitter.com/... @tberzin : No big deal, just automatically segmenting a complex Barrett's image with Meta's new “SAM” tool: https://segment-anything.com/, without any training on Barrett's images. This has the potential to make data annotation and model development dramatically faster.... https://twitter.com/... https://twitter.com/... John Nack / @jnack : Sigh... my @GoogleAI team built something like this five years ago, but I could never get @GooglePhotos to put it to use. Someday, maybe... (Maybe not.) https://twitter.com/... Tanay Jaipuria / @tanayj : Wow! Meta open-sourcing a foundation model for image segmentation (and the dataset!), which can detect even new objects it hasn't seen before. https://twitter.com/... JustDaven / @jayhadhope : “Meta better switch over from VR and MR to AI to catch up” mfs when they realize Meta has been doing AI *just as long and just as well* https://twitter.com/... Fei Xia / @xf1280 : One thing I am particularly excited about this is that SAM has understanding of object parts - which most SoTA open vocabulary segmentation models lack. https://twitter.com/... Darius Chapman / @dariuschapman : Holy joly! This is amazing. We have been trying to build segmentation tool with a specific item. Dumped an image in, and it just worked. 🤯 https://twitter.com/... Mark Tenenholtz / @marktenenholtz : Hey look at that, Meta does actually like open source! Seems to be an extremely powerful segmentation model with a huge dataset also released. Weights + code are licensed under Apache 2.0 (a permissive license). https://twitter.com/... Marco Mascorro / @mascobot : This is a leap in image/pixel segmentation. Meta AI just released SAM (Segment Anything Model). One of the most interesting things is well understating of objects ("objectification" of parts). The model is released open source under an Apache 2.0 license, and it's only 2.4Gb.... https://twitter.com/... https://twitter.com/... Sean Gourley / @sgourley : Huge win in the zero shot computer vision space https://twitter.com/... @drjimfan : Tagging team members to celebrate... Thank you for open-sourcing this wonderful model, while the rest of the AI world is increasingly heading towards the opposite! @drjimfan : Ross (@inkynumbers) was the inventor of Fast R-CNN 7 years ago, which kickstarted CNN-based image segmentation. He co-invented Faster R-CNN and Mask R-CNN. All these years of deep research culminated in Segment-Anything. I have so much respect for Ross and his team. Lior / @alphasignalai : The model was decoupled into: 1) a one-time image encoder 2) a lightweight mask decoder that can run in a web-browser in just a few milliseconds per prompt. Paper: https://ai.facebook.com/... https://twitter.com/... Monica Dinculescu / @notwaldorf : This looks pretty cool, and can lead to actual useful tools for artists, like: segment this image, now replace *this* knife block with a salt shaker and swap it with the lemons for a better composition https://twitter.com/... Ben Hammersley / @benhammersley : Ohhh hellooooo https://twitter.com/... Ayush Jaiswal / @aayushjaiswal07 : A race to open source more resources is exactly what we needed :) https://twitter.com/... Yann LeCun / @ylecun : SAM: Segment Anything Model from FAIR. Foundation model for image segmentation. Demo: https://segment-anything.com/ demo Blog: https://ai.facebook.com/... Paper: https://ai.facebook.com/... Code: https://github.com/... Dataset: SA-1B , 11 million image, 1 billion masks https://ai.facebook.com/... https://twitter.com/... Dhruv Batra / @dhruvbatradb : Segment Anything: general-purpose understanding of objects in images. Model+code under a liberal license following FAIR's commitment to open research. No fear-mongering around this being unsafe for the world. Just the steady (yet fascinating) march of scientific progress. https://twitter.com/... Bardia Khosravi / @khosravi_bardia : Give this a try (with some medical images), it is 🤯! I am thinking for a first pass, rough annotation draft this can be used very effectively. Based on the paper, each image can be segmented in 50 milliseconds (once embedded). https://segment-anything.com/ demo https://twitter.com/... Job / @jobvo : Meta publishing their research and output open source is amazing https://twitter.com/... @vjeux : This one is really cool for front-end engineers, the model is running in the browser using wasm and multithreading. Takes less than 100ms to run so we can do it on mouse move. https://twitter.com/... Simon Willison / @simonw : “the Segment Anything Model is available under a permissive open license (Apache 2.0)” https://twitter.com/... Chris Paxton / @chris_j_paxton : Segment anything has learned a general concept for “objectness.” This is huge, a foundation model for image segmentation is itself a huge step towards a foundation model for robotics and embodied ai https://twitter.com/... Hrafn Thorisson / @hrafntho : Great!! Now add this to @RealityLabs' SDKs https://ai.facebook.com/... https://twitter.com/... Building Jarvis / @building_jarvis : This is huge. Should help Tesla with autonomous driving and Optimus. Also project Jarvis with AR goggles. Tesla should buy Magic Leap. The user data from AR goggles will be the best data sets for Optimus. Let's try to minimize simulation training. https://twitter.com/... @metaai : In addition to the new model, we're also releasing the SA-1B dataset, which is 400x larger than any existing segmentation dataset — we hope this work will help accelerate computer vision research and enable entirely new applications. Get the dataset ⬇️ https://github.com/... @chombabupe : I did a quick basic experiment to see if SAM here can segment basic overlapping geometric shapes. I would expect after training on over a billion objects it would have learnt to model one of the concepts of Gestalt grouping, continuity, but it doesn't it. https://twitter.com/... https://twitter.com/...
Context & Ripple Effects
Meta is pairing a general-purpose segmentation model with SA-1B, a dataset of 11 million images and 1 billion masks, and making the model code available. That turns a research release into reusable computer-vision infrastructure rather than a one-off product feature.
The release sits early in Meta's broader computer-vision push: it was followed by I-JEPA’s alternative approach to visual understanding and, later, a fairness benchmark for image and video classifiers, FACET.
First-order effects
- Researchers and developers can use SAM and SA-1B to build or evaluate object-identification and image-segmentation systems without first assembling a comparably large mask dataset.
- Meta gains a wider external testing and developer base around its vision research, while releasing the code and dataset under permissive terms.
Second-order effects
- Computer-vision teams can shift effort from manual mask labeling toward adapting and validating segmentation models, raising pressure on competing model providers to offer similarly usable tooling or data.
- The size and openness of the dataset make data provenance and safety review more consequential: later scrutiny of large image corpora, including reported harmful material in LAION-5B, illustrates why dataset governance becomes part of model adoption.
Third-order effects
- If open releases of models plus large labeled datasets persist, differentiation in computer vision is likely to move from basic access toward deployment quality, specialized data, evaluation, and trust controls.
- The pattern points to AI infrastructure being distributed through reusable research assets, with access terms and dataset stewardship becoming durable competitive and governance questions.
The trend: This is one data point in the shift from closed, task-specific vision systems toward broadly accessible model-and-dataset stacks for computer vision.