Archive

Posts Tagged ‘caffe’

Qualcomm Snapdragon 845 Octa Core Kryo 385 SoC to Power Premium Smartphones, XR Headsets, Windows Laptops

December 7th, 2017 9 comments

Qualcomm Snapdragon 845 processor was expected since May 2017 with four custom Cortex A75 cores, four Cortex A53 cores, Adreno 630 GPU, and X20 LTE modem. with the launch planned for Q1 2018. At least, that what the leaks said.

Qualcomm has now formally launched Snapdragon 845 Mobile Platform and rumors were mostly right, as the the octa-core processor comes with four Kryo 385 Gold cores (custom Cortex A75), four Kryo 385 Silver cores (custom Cortex A55) leveraging DynamIQ technology, an Adreno 630 “Visual Processing System”, and Snapdragon X20 modem supporting LTE Cat18/13.

The processor is said to use more advanced artificial intelligence (AI) allowing what the company calls “extended reality (XR)” applications, and will soon be found in flagship smartphones, XR headsets, mobile PCs, and more.

Qualcomm Snapdragon 845 (SDM845) specifications:

  • Processor
    • 4x Kryo 385 Gold performance cores @ up to 2.80 GHz (custom ARM Cortex A75 cores)
    • 4x Kryo 385 Silver efficiency cores @ up to 1.80 GHz (custom ARM Cortex A55 cores)
    • DynamIQ technology
  • GPU (Visual Processing Subsystem) – Adreno 630 supporting OpenGL ES 3.2, OpenCL 2.0,Vulkan 1.x, DxNext
  • DSP
    • Hexagon 685 with 3rd Gen Vector Extensions, Qualcomm All-Ways Aware Sensor Hub.
    • Supports Snapdragon Neural Processing Engine (NPE) SDK, Caffe, Caffe2, and Tensorflow
  • Memory I/F – LPDDR4x, 4×16 bit up to 1866MHz, 8GB RAM
  • Storage I/F – TBD (Likely UFS 2.1, but maybe UFS 3.0?)
  • Display
    • Up to 4K Ultra HD, 60 FPS, or dual 2400×2400 @ 120 FPS (VR); 10-bit color depth
    • DisplayPort and USB Type-C support
  • Audio
    • Qualcomm Aqstic audio codec and speaker amplifier
    • Qualcomm aptX audio playback with support for aptX Classic and HD
    • Native DSD support, PCM up to 384kHz/32bit
  • Camera
    • Spectra 280 ISP with dual 14-bit ISPs
    • Up to 16 MP dual camera, up to 32 MP single camera
    • Support for 16MP image sensor operating up to 60 frames per second
    • Hybrid Autofocus, Zero Shutter Lag, Multi-frame Noise Reduction (MFNR)
    • Video Capture – Up to 4K @ 60fps HDR (H.265), up to 720p @ 480fps (slow motion)
  • Connectivity
    • Cellular Modem – Snapdragon X20 with peak download speed: 1.2 Gbps (LTE Cat 18), peak upload speed: 150 Mbps (LTE Cat 13)
    • Qualcomm Wi-Fi 802.11ad Multi-gigabit, integrated 802.11ac 2×2 with MU-MIMO, 2.4 GHz, 5 GHz and 60 GHz
    • Qualcomm TrueWireless Bluetooth 5
  • Location – Support for 6 satellite systems: GPS, GLONASS, Beidou, Galileo, QZSS, SBAS; low power geofencing and tracking, sensor-assisted navigation
  • Security – Qualcomm Secure Processing Unit (SPU), Qualcomm Processor Security, Qualcomm Mobile Security, Qualcomm Content Protection
  • Charging – Qualcomm Quick Charge 4/4+ technology
  • Process – 10nm LPP

The company will provide support for Android and Windows operating systems. eXtended Reality (XR) is enabled with features such as room-scale 6DoF with simultaneous localization and mapping (SLAM), advanced visual inertial odometry (VIO), and Adreno Foveation. Maybe I don’t follow the phone market closely enough, but I can’t remember seeing odometry implemented in any other phones, and Adreon Foveation is not quite self-explaining, so the company explains it combines graphics rendering with eye tracking, and directs the highest graphics resources to where you’re physically looking, while using less resources for rendering other areas. This improves the experience, performance, and lower power consumption.

 

Click to Enlarge

Compared to Snapdragon 835, the new processor is said to be around 25 to 30% faster, the Spectra camera and Adreno graphics architectures are claimed to boost power efficiency by up to 30 percent, and the LTE modem is a bit faster (1.2 Gbps/150Mbps vs 1.0 Gbps/150Mbps). Quick Charge 4+ technology should deliver up  to 50 percent charge in 15 minutes. Earlier this year when SD835 was officially launched, there was virtually no mention of artificial intelligence support in mobile APs, but now NNA (Neural Network Accelerator) or NPE (Neural Processing Engine) are part of most high-end mobile processors, which in SD845 appears to be done though the Hexagon 685 DSP. High Dynamic Range (HDR) for video playback and capture is also a novelty in the new Snapdragon processor.

One of the first device powered by Snapdragon 845 will be Xiaomi Mi 7 smartphone, and according to leaks it will come with a 6.1″ display, up to 8GB RAM, dual camera, 3D facial recognition, and more. Further details about the phone are expected for Mobile World Congress 2018. Considering the first Windows 10 laptop based on Snapdragon 835 processor are expected in H1 2018, we may have to wait until the second part of the year for the launch of Snapdragon 845 mobile PCs.

More details may be found on Qualcomm Snapdragon 845 mobile platform product page.

AWS DeepLens is a $249 Deep Learning Video Camera for Developers

November 30th, 2017 4 comments

Amazon Web Services (AWS) has launched Deeplens, the “world’s first deep learning enabled video camera for developers”. Powered by an Intel Atom X5 processor with 8GB, and featuring a 4MP (1080p) camera, the fully programmable system runs Ubuntu 16.04, and is designed expand deep learning skills of developers, with Amazon providing tutorials, code, and pre-trained models.

Click to Enlarge

AWS Deeplens specifications:

  • Camera – 4MP (1080p) camera using MJPEG, H.264 encoding
  • Video Output – micro HDMI port
  • Audio – 3.5mm audio jack, and HDMI audio
  • Connectivity – Dual band WiFi
  • USB – 2x USB 2.0 ports
  • Misc – Power button; camera, WiFi and power status LEDs; reset pinhole
  • Power Supply – TBD
  • Dimensions – 168 x 94 x 47 mm
  • Weight – 296.5 grams

The camera can not only do inference, but also train deep learning models using Amazon infrastructure. Performance wise, the camera can infer 14 images/second on AlexNet, and 5 images/second on ResNet 50 for batch size of 1.

Six projects samples are currently available: object detection, hot dog not hot dog, cat and dog,  activity detection, and face detection. Read that blog post to see how to get started.

But if you want to make your own project, a typical workflow would be as follows:

  • Train a deep learning model using Amazon SageMaker
  • Optimize the trained model to run on the AWS DeepLens edge device
  • Develop an AWS Lambda function to load the model and use to run inference on the video stream
  • Deploy the AWS Lambda function to the AWS DeepLens device using AWS Greengrass
  • Wire the edge AWS Lambda function to the cloud to send commands and receive inference output

This steps are explained in details on Amazon blog.

Click to Enlarge

Intel also published a press release explaining how they are involved in the project:

DeepLens uses Intel-optimized deep learning software tools and libraries (including the Intel Compute Library for Deep Neural Networks, Intel clDNN) to run real-time computer vision models directly on the device for reduced cost and real-time responsiveness.

Developers can start designing and creating AI and machine learning products in a matter of minutes using the preconfigured frameworks already on the device. Apache MXNet is supported today, and Tensorflow and Caffe2 will be supported in 2018’s first quarter.

AWS DeepLens can be pre-ordered today for $249 by US customers only (or those using a forwarding service) with shipping expected on April 14, 2018. Visit the product page on AWS for more details.

Intel Introduces Movidius Myriad X Vision Processing Unit with Dedicated Neural Compute Engine

August 29th, 2017 No comments

Intel has just announced the third generation of Movidius Video Processing Units (VPU) with Myriad X VPU, which the company claims is the world’s first SoC shipping with a dedicated Neural Compute Engine for accelerating deep learning inferences at the edge, and giving devices the ability to see, understand and react to their environments in real time.

Movidius Myraid X VPU key features:

  • Neural Compute Engine – Dedicated on-chip accelerator for deep neural networks delivering over 1 trillion operations per second of DNN inferencing performance (based on peak floating-point computational throughput).
  • 16x programmable 128-bit VLIW Vector Processors (SHAVE cores) optimized for computer vision workloads.
  • 16x configurable MIPI Lanes – Connect up to 8 HD resolution RGB cameras for up to 700 million pixels per second of image signal processing throughput.
  • 20x vision hardware accelerators to perform tasks such as optical flow and stereo depth.
  • On-chip Memory – 2.5 MB homogeneous memory with up to 450 GB per second of internal bandwidth
  • Interfaces – PCIe Gen 3, USB 3.1
  • Packages
    • MA2085: No memory in-package; interfaces to external memory
    • MA2485: 4 Gbit LPDDR4 memory in-package

The hardware accelerators allows to offload the neural compute engine, for example, the stereo depth accelerator can simultaneously process 6 camera inputs (3 stereo pairs) each running 720p resolution at 60 Hz frame rate. The slide below also indicates Myriad X to have 10x higher DNN performance compared to Myriad 2 VPU found in Movidius Neural Compute Stick.

Click to Enlarge

The VPU ships with an SDK that contains software development frameworks, tools, drivers and libraries to implement artificial intelligence applications, such as a specialized “FLIC framework with a plug-in approach to developing application pipelines including image processing, computer vision, and deep learning”, and a neural network compiler to port neural networks from Caffe, Tensorflow, and others.

Myriad SDK Architecture

More details can be found on Movidius’ MyriadX product page.

Autonomous Deep Learning Robot Features Nvidia Jetson TK1 Board, a 3D Camera, and More

January 25th, 2016 No comments

Autonomous, a US company that makes smart products such as smart desks, virtual reality kits and autonomous robots, has recently introduced a deep learning robot that comes with a 3D camera, speaker and microphone, Jetson TK1 board, and a mobile base.

Autonomous_Deep_Learning_Robot

The robot appears to be mostly made of the shelves parts:

  • 3D Depth camera – Asus Xtion Pro 3D Depth Camera
  • Speaker & Microphone
  • Nvidia Jetson TK1 PM375 board – Nvidia Terra K1 quad-core Cortex A15 processor @ 2.3 GHz with a 192-core Kepler GPU, 2GB RAM, 16 GB flash
  • Kobuki Mobile Base –  Kobuki is the best mobile base designed for education and research on state of the art robotics. Kobuki provides power supplies for external computer power as well as additional sensors and actuators. Its highly accurate odometry, amended by calibrated gyroscope, enables precise navigation.

The robot is designed for research in deep learning and mobile robotics, and comes with Ubuntu, Caffe, Torch, Theano, cuDNN v2, and CUDA 7.0, as Robot Operating System (ROS) set of open source software libraries and tools.

Kobuki Base

Kobuki Base

While there’s virtually no documentation at all on the product page, I’ve been told that the robot was built on top of TurtleBot open source robot, and re-directed to tutorials available via TurtleBot Wiki, as well useful resources for deep learnings frameworks such as Caffe and Torch, and Google TensorFlow Tutorials.

Autonomous Deep Learning Robot sells for $999 with manual charging, or $1048 with a self-charging dock.

Thanks to Nanik for the tip!