Disney Research Studios introduces FRAN, a cutting-edge face ageing network simplifying actor ageing in videos. Photorealistic digital re-aging, while common, has been labour-intensive, often taking weeks of manual work by artists. To tackle this, Disney researchers from Zürich generated thousands of fictional faces, training the U-Net architecture to predict age-related changes and automatically add or remove wrinkles. This AI technology streamlines the ageing process, sparing Hollywood studios from costly reshoots. By layering the details onto the original face, FRAN retains the actor's authentic appearance while allowing artists to fine-tune the ageing effect with localised control and creative freedom, marking a significant advancement in filmmaking technology.
In the realm of board games, artificial intelligence has achieved remarkable feats in chess and Go, surpassing human capability long ago. However, more complex games like Stratego and Diplomacy have presented challenges for computer systems until recently. AI has made impressive progress in Stratego, becoming highly skilled at this game of strategy based on imperfect information. Diplomacy, known for its focus on alliances and secret agreements, has also seen notable advancements in AI play. The key difference in these games lies in their reliance on hidden information, requiring the AI to make educated guesses about the opponent's intentions and potential moves. Soft skills such as bluffing and convincing opponents play a significant role, making it more than just a matter of raw calculation. AI models like DeepNash from DeepMind have demonstrated success by strategically sacrificing pieces and anticipating their opponents' reactions to outmanoeuvre them.
Baidu, often referred to as the Google of China, reveals its extensive robotaxi plans, with ambitions to create the world's largest no-safety-driver robotaxi service area by 2023. Already operational in Chongqing and Wuhan, Baidu's robotaxi fleet has provided close to 1.4 million rides, with nearly half a million in Q3 of 2022. The company has developed its custom AI hardware and launched a subsidiary to produce its AI accelerator, enabling the creation of larger and more sophisticated models, including end-to-end machine learning systems for generating driving commands. Baidu's ‘Big Model’ aims to tackle rare and novel driving scenarios, addressing the challenges of the long tail of corner cases in self-driving technology, and the company boasts success in recognising thousands of objects and handling unexpected situations and obstacles.
Waabi, the Toronto-based autonomous driving startup, rolls out its first robotic semi-trucks for on-road testing in the U.S. only a year after emerging from stealth mode, and its CEO Raquel Urtasun believes they can commercialise the technology faster than their better-funded predecessors. Unlike larger AV companies such as Waymo and Cruise, Waabi positions itself as a second-generation AV company, emphasising aggressive virtual testing through its ‘Waabi World’ simulator, which enables the AI system to simulate a wide range of road conditions and scenarios, accelerating the learning process. With an $83.5m funding round, Waabi aims to revolutionise self-driving technology with its innovative approach, potentially disrupting the autonomous vehicle landscape.
Despite the challenges, augmented reality (AR) holds vast potential for commercial and consumer applications, with experts suggesting it could replace smartphones in the future. However, realising AR's potential requires advancements in AI technology, usage models, and business strategies that may take time to mature and become affordable for the masses. Tirias Research estimates this timeline to be around five to ten years. Recently, Qualcomm made a noteworthy stride in bringing this vision closer to reality by unveiling its first chipset platform tailored specifically for AR glasses. The Snapdragon AR2 platform intelligently partitions silicon blocks into three modules strategically positioned around the glasses, enabling sleeker and more comfortable designs while meeting the high compute power and connectivity demands. This progress paves the way for lighter and more practical AR glasses with realistic performance expectations.
Back to Blogs