However,
Meta unveils dinov3, self-supervised ai:
Dinov3, a new Meta IA model, learned to interpret images without human annotations. Similarly, It can be used for a wide range of applications, ranging from embedded systems to environmental surveillance.
Meta unveiled Dinov3, a computer vision AI model, which has the distinction of being from self-supervised learning (self-supervised learning, SSL). In addition, Depending on the company, this version is able to compete, or even exceed, traditional supervised systems on many tasks. For example, This new system would be efficient on images from the web as on shots taken by satellites.
Unlike conventional approaches based on large sets of manually annotated images, Dinov3 learns without labels. Similarly, The model was drawn over 1.7 billion images and has up to 7 billion parameters. Similarly, Meta claims that this method strongly reduces the time and costs necessary for training.
Diversified capacities – Meta unveils dinov3, self-supervised ai
Dinov3’s abilities include classification of meta unveils dinov3, self-supervised ai images. semantic segmentation (identify and delimit objects in an image) and monitoring objects in a video. Dinov3 is also distinguished by its ability to produce visual representations in very high resolution. These representations. which finely describe the content of the image, can be used by small additional modules, requiring few annotated examples to obtain efficient results. The same basic model can thus be used. without heavy modifications, for several parallel applications, optimizing calculation costs, in particular for on -board systems or IT environments with limited resources.
Meta offers in addition to the most powerful version. a family of more compact models – lives and Prévnese – intended for researchers and developers with limited resources. All are made available with the training code and pre-trained models, under commercial license.
Meta underlines that the versatility of Dinov3 opens up perspectives in various sectors. from health to automotive, including surveillance or meta unveils dinov3, self-supervised ai logistics. In the environmental field. Meta cites the example of the World Resources Institute (WRI), which already uses Dinov3 to analyze satellite images and measure the height of the canopy in Kenya. The precision obtained made it possible to reduce the average error by 4.1 meters to 1.2 meters, in particular facilitating the automation of monitoring of reforestation projects.
Further reading: It had to happen: Grok will land in the Tesla – The pressure is increasingly intensifying – The EU defends its commercial agreement with Trump – Small practical guides for hunters and fishermen – The New York Stock Exchange clings to the prospects for lowering the rates of the Fed.
