Case Study: Harnessing Computer Vision and Machine Learning for Real-World Products

Submitted by BDTI on Thu, 11/16/2017 - 01:00

The buzz around artificial intelligence, computer vision and machine learning intensifies on a daily basis: There are a dizzying number of new processors, algorithms and tools for computer vision and machine learning. Investment and acquisition activity around AI companies is furious. Announcements of new AI-based applications and products are non-stop. Competition for engineering talent is fierce.

All this creates challenges for product designers and application developers who seek to leverage computer vision and machine learning. Which processing platform will provide the best mix of performance, cost and power consumption for my application? What tools and libraries are available? Which software framework should I adopt? How much and what types of training data do I need? And, can I even solve this problem using computer vision and machine learning?

BDTI helps companies effectively use computer vision and machine learning. For 25 years, BDTI has been analyzing processors, algorithms and tools for embedded processing applications. And, we’ve been building embedded applications, too. We began focusing on computer vision and machine learning about 7 years ago, and since then we’ve have completed dozens of evaluation and development projects for clients ranging from start-ups to Fortune 500 companies. BDTI’s unique combination of industry-wide perspective and hands-on technical experience in application development enables us help companies make well-informed strategic decisions, identify best-fit technology, and design and build winning products.

For example, when a leading venture capital firm needed to understand the commercial viability of a very complex new product in order to decide whether or not to invest in its manufacturer, the VC firm turned to BDTI. The investors were attracted by the market potential and leading-edge technology of the product, but didn't have the technical knowledge to evaluate it. For this, they engaged BDTI. BDTI carefully examined what was under the hood, assessing whether the developers would be able to deliver the promised features and performance. BDTI's detailed observations enabled the investor to make a well-informed investment decision.

In another engagement, the R&D group of a worldwide manufacturer of automotive technology asked BDTI to survey processors for deep learning in order to save time and reduce risk in its evaluation process. BDTI engineers, experienced in the uses and nuances of deep learning technologies, conducted a survey of dozens of processor options. The report provided the company with valuable insights—identifying new potential suppliers and also uncovering processor features, supported deep learning algorithms and frameworks, tools, and other key details. With this information, the company was able to move quickly to begin its own evaluation of the most promising candidates for its applications.

In several customer engagements, BDTI created computationally efficient deep learning algorithms running on embedded platforms. For one client, BDTI used a convolutional neural network to identify objects and their locations in video frames on an embedded processor. For this, BDTI first performed a detailed estimation of the processing performance and memory bandwidth required for each layer in the network, determining the optimal number of layers needed to obtain sufficient accuracy and efficient use of processor resources. Next, BDTI created a reference implementation and converted it to 8-bit data types to ensure efficient execution. Finally, BDTI optimized several key functions, and then iterated training and fine-tuning. The result was a highly accurate and efficient network.

In each of these engagements, BDTI’s clients gained competitive advantage through BDTI’s knowledge and expertise in computer vision and deep learning. To learn how BDTI can help your company, contact Jeremy Giddings by phone at +1 (925) 954 1411 or via email.

Add new comment

Log in to post comments