Embedded World Conference 🎙 Speaker: Martin Croome, VP Marketing, GreenWaves 📢 Presentation: "GAP9: A RISC-V Based NN/DSP Application Processor Enabling Next Generation Audio Features in Hearable and Wearable Products" 🗓 Date: Thursday, April 11 🕤 Time: 9:30 AM 📍 Location: NCC OST CONVENTION CENTRE Please submit this form if you would [...]
Please submit this form if you would like to meet us at CES 24
GreenWaves Technologies announces a €20M financing The funding will support the production ramp up of GAP9 and the development of its next generation GAP processor. GreenWaves, the French pioneer in RISC-V application processors for battery powered devices and [...]
GreenWaves will demonstrate live the ground-breaking AI and DSP demos on its ultra low power chip at Embedded World 2022 Located in the micro-and nano-electronics hotspot of Grenoble, GreenWaves is a fabless semiconductor startup that designs and brings to market advanced ultra-low-power AI and DSP processors for energy-constrained applications. Our [...]
GreenWaves will present at Embedded World 2021 Greenwaves Technologies will present its groundbreaking technologies at Embedded World 2021. We will demonstrate GAP IoT application processors based on RISC-V architecture and will share our knowledge and expertise in Edge Computing and Artificial intelligence. Loic Lietar, co-founder and CEO of GreenWaves [...]
自动驾驶玩具车成为可能 http://player.youku.com/embed/XNDQ2NTc1MDQxNg
基于GAP8的毫瓦级人脸识别 https://player.youku.com/embed/XNDQ2NTUzOTA2NA 人脸识别因为在安全领域被采用而吸引了媒体的大量关注。然而人脸识别其实是一个可以被广泛应用的有趣特性。识别设备的拥有者（区别于其他使用者）可以创造出不一样的用户体验。是否有方案可以增加人脸识别功能；而仅增加几美元的成本；并适合在电池供电下运行数年？GAP8让设想成为可能！ GreenWaves发布了一个完整软件应用（可运行在GAPPoC机器视觉参考设计板上）。人脸识别应用使用人脸检测和人脸识别算法：人脸检测采用的是Viola-Jones经典人脸检测算法，人脸识别采用基于SqueezeNet卷积神经网络。此应用完美地演示了GAP8在极低功耗下可以灵活地运行不同的算法。 首先，这个应用采用人脸检测算法在图像中寻找人脸。直到检测到一张人脸后，人脸识别算法才会启动。人脸检测如采用例如被动红外探测器作为外部触发信号，这样可以进一步减小无人情况下的功耗。当人脸被检测到，人脸检测算法会输出人脸在图像中的坐标信息。提取这个区域并置于28像素×28像素图像中。此提取图像作为人脸识别算法的输入，而算法的输出是被检测到人脸的1024个（16比特）特征参数。 已被识别人脸的特征参数被存储在数据库中。同一张脸再一次经过人脸识别算法网络会产生一套与已存储特征参数高度相关的参数。所以机器学习一张或者多张图像，然后将这些特征参数结果存入到数据库中。一旦经过人脸识别网络产生的特征结果和数据库中特征相比较非常接近，一次匹配就完成了。 输入的图像分辨率是QVGA（320*240像素）。人脸检测算法功耗约1毫瓦/帧，每秒可处理约70帧。 人脸识别卷积神经网络功耗约22毫瓦/帧/秒。此神经网络只有在人脸被检测到后才会启动，约需要400毫秒完成识别。此神经网络在Labelled Faces in the Wild dataset (LFW)数据集上达到了96%的准确率。(Link) 在此功耗级别下，人脸识别功能可以被整合到大量不同设备中而对设备的电池寿命影响很微小。 GreenWaves正式发布了整个人脸识别应用，包括训练脚本、基于BSD开源许可的GAP AutoTiler模型。可访问如下链接获取全部资料： https://github.com/GreenWaves-Technologies/FaceReID
