Would deep learning work for your application?
Do you need a deep learning algorithm designed and implemented?
Want to get more performance out of your deep learning algorithm?
Contact BDTI to discuss your needs.
- Yann LeCun on "Convolutional Neural Networks" at the 2014 Embedded Vision Summit
- "Efficient Implementation of Convolutional Neural Networks using OpenCL on FPGAs" at the 2015 Embedded Vision Summit
- "Tailoring Convolutional Neural Networks for Low-Cost, Low-Power Implementation" at the 2015 Embedded Vision Summit
- Jeff Bier includes discussion of Synopsys' new processor for CNNs in "Choosing a Processor for Embedded Vision: Options and Trends"
- InsideDSP article: "Neural Network Processors: Has Their Time Come?"
- InsideDSP article: "Here Come the Learning Machines"
- InsideDSP article: "Creating Machines That See: Show or Tell?"
Deep Learning for Computer Vision
Deep learning—frequently implemented using artificial neural networks (ANNs) such as convolutional neural networks (CNNs) and deep neural networks (DNNs)—has emerged as a compelling approach for a range of computer vision tasks. Deep learning models human perception and has proven highly successful in a range of applications, including computer vision.
BDTI has hands-on experience with CNNs for vision applications on a wide variety of processing platforms and can help your company select the most suitable processor for your CNN-based application. BDTI's work with several silicon and silicon IP vendors highlights this experience. For example:
A leading electronic design tools company engaged BDTI to analyze its silicon IP for implementing CNN-based applications in low-cost, low-power products. Relying on long experience in developing embedded signal processing applications on a wide range of processing platforms, BDTI carefully examined the architectural design and implementation methodology through hands-on analysis. The results of the analysis showed that the architecture has an attractive performance-power-area trade-off compared to GPUs.
NVIDIA engaged BDTI to evaluate the Jetson TX1 Developer Kit by developing a deep-learning-based computer vision application. BDTI also used the kit to develop a classic computer vision algorithm. Using its knowledge of embedded application development practices, BDTI was able to identify several stand-out features of the kit.
A programmable logic vendor, seeking new markets for its devices, engaged BDTI to evaluate the suitability of the company's architecture for CNN-based vision applications. By relying on its years of designing and implementing benchmarking methodologies, BDTI developed an approach involving implementation of a convolutional layer on the customer's architecture and a competitor. The results pointed out several unique advantages of programmable logic over GPUs.
Or, if you need a CNN for your product, BDTI can design and build a CNN-based application for you. Also, if you need better performance from your existing CNN-based application, BDTI can optimize its performance. Customer engagements include:
A programmable processor vendor engaged BDTI to build a CNN-based pedestrian detection application that would run efficiently on its ARM-based SoC. BDTI selected a suitable CNN architecture, then with the target processor in mind, experimented with various designs to reduce computational demand while maintaining acceptable accuracy. BDTI's implementation efficiently and reliably runs on the customer's programmable processor.
Call BDTI at +1 925 954 1411 or contact us via the web to discuss how BDTI can help with your computer vision project. All inquiries are confidential.