Maarten Grootendorst's new book, the beautiful "Segment Anything 2" (SAM 2) for mac and the Vchitect-2.0 engine

Here are the news of the day...

Maarten Grootendorst's new book is out, it is called "Hands-On Large Language Models" and I have already ordered it (with 250 color visuals!) :)

Amazon (no referral)

The AI research team at Meta has unveiled Segment Anything 2 (SAM 2), a cutting-edge computer vision model for image and video segmentation. This new version offers significantly improved performance compared to its predecessor, with extremely fast inference times on devices like Macs and iPhones thanks to optimization for Apple's CoreML.

One of the key features of SAM 2 is its ability to perform inference in a fraction of a second, enabling near-instantaneous annotation of any image. The team has released optimized model checkpoints in various sizes (tiny, small, base, and large) under the Apache license, allowing developers to choose the best trade-off between size and performance based on their needs.

In addition to the open-source image annotation application, AIatMeta is working on adding video support in upcoming releases. Guides for fine-tuning SAM 2 for specific use cases, such as Medical SAM for medical applications, are also available.

To facilitate the adoption of this new technology, the team has released a ready-to-use packaged app to run SAM 2 directly on Macs. Users can download and try it immediately to experience the capabilities of this revolutionary computer vision model.

While the advancements in SAM 2 are impressive, it's important to note that like any AI system, it may have limitations or biases that should be carefully considered and addressed. Responsible development and deployment of such technologies are crucial to ensure they benefit society while mitigating potential risks or unintended consequences.

You can download it here.

Then this new text-to-image engine called Vchitect-2.0 came out.
It looks interesting, there is even a demo on HF that generates a few seconds.