UAE's TII Introduces Falcon Perception AI Model to Compete Globally

The Technology Innovation Institute (TII), part of Abu Dhabi's Advanced Technology Research Council (ATRC), has proudly introduced Falcon Perception, a cutting-edge multimodal AI model tailored to compete with top global technologies.

Featuring about 600 million parameters, Falcon Perception is compact yet powerful. It excels in various tasks such as object segmentation, visual comprehension, and document analysis, demonstrating capabilities similar to those of significantly larger AI frameworks while requiring notably less computational power.

Multimodal AI encompasses systems capable of processing multiple data types simultaneously—like images and text. Contrary to traditional models that primarily address language, Falcon Perception aims to integrate visual understanding with text interpretation seamlessly.

According to TII, this innovative model can detect and segment objects in intricate images, extract text from documents, and engage in natural language queries about visual content. For instance, users can request the model to spot specific items in busy visuals or tally objects within a scene, and it reliably highlights and locates them.

By consolidating visual and linguistic processing into one system, this design minimizes the necessity for multiple systems, enhancing ease of use and efficiency, particularly beneficial in resource-limited real-world applications.

Potential applications for Falcon Perception include robotics, industrial automation, manufacturing inspections, and extensive data labeling for AI training.

Dr. Najwa Aaraj, CEO of TII, emphasized that this model underscores the institute's commitment to creating pragmatic and sophisticated AI systems deployable across various sectors, thereby fortifying the UAE's sovereign AI capabilities.

Benchmarks show that Falcon Perception excels in critical areas such as object segmentation, advanced visual interpretation, and document intelligence, performing comparably to larger models like Meta's SAM3 and Alibaba's Qwen in crucial evaluation metrics.

Dr. Hakim Hacid, Chief Researcher at TII’s Artificial Intelligence and Digital Research Center, noted that this model exemplifies how a single efficient architecture can adeptly manage complex vision and language tasks without relying on multi-stage systems.

As the first Falcon initiative focusing on dense multimodal perception, Falcon Perception is set to be available as open-source on Hugging Face, fostering global collaboration and encouraging further advancements in multimodal AI technologies.