#054 - Evolution of Computer Vision: From Classical Algorithms to Deep Learning Networks, AMD Acquires Nod.ai to Boost its AI Software Ecosystem, Foxconn & Nvidia Partner to Accelerate EV Development
Fresh & Hot curated AI happenings in one snack. Never miss a byte 🍔
This snack byte will take approx 6 minutes to consume.
AI BYTE # 1 📢 : The Evolution of Computer Vision: From Classical Algorithms to Deep Learning Networks
⭐ Computer vision (CV) is a fascinating field that has been evolving for decades, from the early studies in the 1970s to the recent breakthroughs in deep learning.
Deep Learning has transformed many CV problems, such as object detection, feature extraction, and semantic segmentation. However, Deep Learning is not a silver bullet that can solve every CV challenge.
In fact, there are still some tasks that are better suited for classical CV algorithms, such as Simultaneous Localization and Mapping (SLAM) and Structure From Motion (SFM).
In this post, I will share some insights on how Deep Learning and classical CV coexist and complement each other, and why we should not discard the old methods in favor of the new ones.
Deep learning is a form of AI that uses neural networks to learn from data and solve complex problems. It has shown remarkable results in many CV tasks, especially when paired with large labeled image databases.
For example, Convolutional Neural Networks (CNNs) and Region-based CNNs (R-CNNs) have made object detection much easier and more accurate than before, without requiring explicit rules or sliding windows.
Similarly, CNNs and U-net architectures have simplified feature extraction and semantic segmentation, eliminating the need for handcrafted methodologies or region separation.
However, Deep Learning is not a magic solution that can handle every CV problem. It has some limitations and drawbacks, such as requiring a lot of data and computational power, being prone to overfitting or underfitting, and lacking interpretability or explainability.
Moreover, Deep Learning is not very effective when it comes to problems that involve complex mathematics or geometry, such as SLAM and SFM.
SLAM is a technique that allows an agent (such as a robot or a car) to build and update a map of an environment while keeping track of its location within the map. This is essential for autonomous navigation and exploration.
SFM is a technique that allows us to create a 3D reconstruction of an object or a scene using multiple images taken from different viewpoints. This is useful for applications such as 3D modeling, virtual reality, or augmented reality.
Both SLAM and SFM rely on classical CV algorithms that use advanced mathematics and geometry to estimate the camera pose, the 3D structure, and the motion of the scene.
These algorithms are based on close approximations that make the computational requirements more manageable. They also use only the camera’s intrinsic properties and the features of the image, which makes them more cost-effective than other methods such as laser scanning.
These classical CV algorithms have proven to be reliable and accurate in solving SLAM and SFM problems, while Deep Learning approaches have not been able to match their performance or efficiency. Therefore, classical CV still dominates these specific challenges.
The lesson here is that we should not blindly replace classical CV with deep learning, but rather identify which problems are best solved by which techniques.
We should also appreciate the artistry and creativity involved in classical CV methods, which require us to formulate and solve mathematical problems rather than rely on data-driven learning.
I believe that the future of CV will not be about learning alone, but also about understanding. We should aim to develop networks that can comprehend information deeply and reach meaningful conclusions with minimal intervention.
We should also seek to integrate classical CV algorithms with Deep Learning networks to leverage their strengths and overcome their weaknesses.
I hope you enjoyed this post and learned something new about CV. If you have any questions or comments, please feel free to share them below. Thank you for reading!
AI BYTE # 2 📢 : AMD Acquires Nod.ai to Boost its AI Software Ecosystem
⭐ AMD has announced plans to acquire Nod.ai, a startup that specializes in optimizing AI software for high-performance hardware.
This acquisition will significantly enhance AMD’s ability to provide AI customers with open software that allows them to easily deploy highly performant AI models tuned for AMD hardware.
Nod.ai is a startup that provides key enabling technologies for future AI systems using advanced compiler-based approaches, instead of legacy handwritten kernels.
The company created the SHARK Machine Learning Distribution, which is built on LLVM, MLIR, OpenXLA’s IREE and Nod.ai’s tuning.
Nod.ai’s software can accelerate the deployment of AI models across a broad range of platforms powered by AMD’s architectures, such as Instinct data center accelerators, Ryzen AI processors, EPYC processors, Versal SoCs and Radeon GPUs.
The acquisition of Nod.ai is expected to significantly enhance our ability to provide AI customers with open software that allows them to easily deploy highly performant AI models tuned for AMD hardware, according to Vamsi Boppana, Senior Vice President, Artificial Intelligence Group at AMD.
The acquisition also underscores AMD’s growth strategy in the AI sector, which is centered on an open software ecosystem that simplifies the adoption process for customers through developer tools, libraries, and models.
This acquisition adds another feather to AMD’s cap as it continues its expansion into the rapidly evolving AI industry.
AMD has been investing heavily in AI technologies in recent years , such as CDNA, XDNA, RDNA and Zen architectures, to compete with rivals like Nvidia and Intel in the fast-growing AI market.
According to an industry report, the global AI market size is estimated to reach around $594 billion by 2032.
In addition to the technology, AMD said it aims to leverage the engineering talent from Nod.ai to boost its open-source developer cred. Nod.ai is a contributor to AI software repositories like SHARK and Torch-MLIR used by many researchers.
AI BYTE # 3 📢 - Foxconn and Nvidia Partner to Accelerate EV Development with Powerful AI Solutions
⭐ The EV industry is undergoing a rapid transformation, driven by the advances in AI. It enables EVs to achieve higher levels of automation, safety, and efficiency, creating a better driving experience for users.
One of the key players in this field is Foxconn, the world’s largest electronics manufacturer. Foxconn has recently announced a strategic partnership with Nvidia, the leading provider of AI solutions for automotive applications.
The partnership aims to leverage Nvidia’s comprehensive suite of automotive products, including the Nvidia Drive Hyperion 9 platform, the Drive Thor Central Computer, and an Advanced Sensor Architecture.
Foxconn’s vision is to create software-defined vehicles with centralized electronic architectures that can handle the immense computational demands of highly automated and self-driving vehicles.
To achieve this, Foxconn will use Nvidia’s Drive Orin platform, which has been chosen by over 25 global automakers, as the AI brain for its EVs.
Foxconn will also produce Electronic Control Units (ECUs) featuring the upcoming Drive Thor superchip, which is expected to deliver 2,000 teraflops of high-performance compute power.
The collaboration will also leverage the Drive Hyperion 9 platform, a modular development platform for automated and autonomous vehicles.
Powered by Drive Thor, it integrates a qualified sensor architecture capable of level 3 urban and level 4 highway driving scenarios. With a combination of high-resolution cameras, radar, lidar, and ultrasonic sensors, Drive Hyperion processes a vast amount of safety-critical data to enable precise navigation.
One of the key advantages of Drive Hyperion is its compatibility across generations, ensuring a seamless transition from Drive Orin to Drive Thor and beyond.
This compatibility, along with Nvidia’s stringent qualification processes for sensors, helps streamline development time and reduce costs for manufacturers like Foxconn.
The partnership between Foxconn and Nvidia marks a significant step towards realizing the potential of AI for EVs.
By combining their expertise and resources, they can create innovative and intelligent EVs that will transform the future of mobility.