On April 5, Meta, the parent company of Facebook, announced on Twitter that the SAM model (Segment Anything Model) was launched and open sourced. The model is capable of finding and segmenting any object in images and videos, enabling one-click matting. On the same day, Meta also released what it claims is the largest image annotation dataset ever, as well as a 30-page detailed paper "Segment Anything".
1.1 billion+ masked datasets to extract!
The goal of SAM is to accurately "segregate everything". Compared with the widely visible manual cutout function, SAM has the following outstanding highlights in the operation level:
first, everything recognition. "Zero-sample performance" is its strength. In the face of unseen objects, SAM can recognize it without additional training.
Second, it can handle multi-modal prompts such as text, key points, bounding boxes, etc. If you input specific text such as "CAT", SAM can segment objects in the picture.Third, given any picture, SAM can automatically segment all the items in the image and classify them one by one.
Fourth, in terms of video, SAM can also accurately identify and quickly mark, and automatically use ID to record and classify these items.
Fifth, SAM accepts input prompts from other systems and outputs object information to other AI systems.
The Meta R&D team released the details of the paper on Segment Anything on the official website.
Dr. Tang Yudi also explained the paper and code with video
If you need videos and paper codes, you can pay attention to the public H: Gupao AI Reply 211 Get it
Free prostitution 60G entry to advanced AI resource package (classic papers/books/project courses/learning route outline) + with papers/study/employment/competition guidance + Daniel technical questions and answers
Included: deep learning neural network + CV computer vision Learning (two major frameworks pytorch/tensorflow + source code courseware notes) + NLP, etc.
Applicable people
① Students who are preparing for graduation
thesis ② AI algorithm engineers who are planning to change jobs and are looking for a job, etc.
③ Self-study and people who are going to switch to the AI field
④ Want to consolidate the core of AI Knowledge, people who check for gaps