CEVA-XM4 - Intelligent Vision Processor
Bringing Human-Like Intelligent Vision Processing to Low-Power Embedded Systems
CEVA-XM4 is the 4th generation imaging and computer vision processor IP from CEVA, designed to bring human-like vision and visual perception to embedded systems.
CEVA-XM4 solves the most critical issues for the development of energy-efficient embedded vision systems where the die size and power budget are extremely constrained, yet the algorithms require heavy processing capabilities.
CEVA-XM4 enables the three key elements of intelligent vision processing:
3D vision, Computational Photography, Visual Perception and Analytics
Imitating human vision and adding computational photography to embedded systems is challenging, given the typically limited space and power resources. To address these issues, CEVA incorporated programmable wide-vector architecture, with fixed- and floating-point processing, multiple simultaneous scalar units, and a vision-oriented low-power instruction set. As a result, the CEVA-XM4 achieves up to 8x performance improvement with 2.5 greater energy efficiency, compared to the CEVA-MM3101.
The CEVA-XM4 outperforms the most advanced GPU cluster shipping today by 33% faster processing, while consuming 9x less energy and is 17x smaller in die size.
With the CEVA-XM4 you can
- Run computer vision on video streams (1080p, 4K)
- Combine depth generation with vision processing (e.g. depth + augmented reality, depth + 3D scanning)
- Enable multi-app processing (e.g. gesture + face detection + emotions + eye-tracking + optional depth)
- Implement multi-frame algorithms in high resolution (e.g. super-resolution, refocus)
For a full list of features and capabilities, download the CEVA-XM4 white paper.
CEVA Deep Neural Network (CDNN)
The CDNN is a real-time software framework for CEVA-XM4 to implement deep learning for object recognition and vision analytic algorithms.
CDNN includes 2 major components:
- CEVA Network Generator – an offline PC tool that converts existing pre-trained floating point networks into fixed-point networks which are optimized for low-power
- CNN (convolutional neural network) real-time libraries and APIs – CEVA-XM4 optimized software libraries covering any type or portion (layer) of a network including various APIs (to the full network or to the portion/layers of the network). It is modular and optimized for low memory bandwidth.
The CEVA Super-Resolution (SR) algorithm increases quality of still images for mobile devices. In today’s world, where smartphones are the most commonly used camera devices, image quality is still limited due to slim form factor and strict power constraints. Super-Resolution is an important breakthrough, enabling systems manufacturers and OEMs to increase the quality of images on any device, regardless of these factors.
Digital Video Stabilizer
The CEVA Digital Video Stabilizer (DVS) is an advanced software application that is able to handle both shakiness (in multiple axes) and rolling shutter effects to output smoother video streams in any light conditions for embedded devices such as smartphone, tablets and others.
The CEVA-XM4 is an extremely high performance, fully programmable, low power, fully-synthesizable Digital Signal Processor (DSP) and memory subsystem IP core.
The processor key features include:
- Fully programmable in high level languages
- Scalar and Vector units to handle a mix of control and parallel code efficiently
- Very Long Instruction Word (VLIW) and Single Instruction Multiple Data (SIMD) functionality
- Full memory sub-system for easy integration into SOCs, utilize multi-core and hardware accelerator connectivity using standard interfaces
- Automated traffic management from the system into local memories to best performance and power efficiency
- Flexible precision: A combination of mixing efficient fixed-point and floating-point math
CEVA-XM4 comes with a complete development platform including
- Extensive vision libraries accessible directly from the CPU for offloading the computer vision functions to achieve energy saving
- Robust software development tools and a software development framework
- Hardware development platform
- Product level software applications developed by CEVA, including Digital Video Stabilizer and Super Resolution
CEVA-XM4 targets any camera-enabled devices such as smartphones, tablets, automotive safety (ADAS) and infotainment, robotics, security and surveillance, augmented reality (AR) and virtual reality (VR), drones, and signage.
Areas of expertise relate to quality image preparation for 3D vision, image enhancement supporting various computational photography applications and generating sophisticated visual perception and video analytics from the input data.
Computer vision algorithms supported include:
- Real-time 3D depth map generation
- Point cloud processing
- 3D scanning
- Object detection and tracking for example used in augmented reality (AR) applications
- Object, image and scene recognition
- Natural user interface applications such as face recognition, gesture recognition, user intent and emotion recognition user authentication and the like
Computational Photography algorithms supported include
- Refocus image
- Background replacement
- Image and video stabilization
- Noise reduction
- Improved low-light capabilities.
CEVA-XM Announcements16 Aug 2016 Novatek Licenses CEVA-XM4 Imaging and Vision DSP for Embedded Visual Intelligence
10 Aug 2016 Rockchip and CEVA Extend Partnership to Include CEVA-XM4 Intelligent Vision DSP
27 Jun 2016 CEVA's 2nd Generation Neural Network Software Framework Extends Support for Artificial Intelligence Including Google's TensorFlow
- Products Overview
- DSP Cores
- Connectivity Platforms
- Development Environment