首款开源软硬一体OpenCV AI Kit(OAK)用户问题官方答疑实录 Q&A

作为计算机视觉领域的老牌军,OpenCV 涵盖图像处理和计算机视觉方面的很多通用算法,是非常有力的研究工具之一,且稳居开发者最喜爱的 AI 工具/框架榜首。为了让大家有机会得以上手实践,CSDN 于上周六特邀OpenCV CEO Satya Mallick、OpenCV中国团队核心成员吴佳以及OpenCV AI Kit 开发工程师 Steven Li 做客CSDN 直播间,为大家分享OpenCV的最新技术进度、 OpenCV AI Kit (OAK)主要功能及应用场景,点此可观看直播回放



a. No.  The Myriad X is listed for sale and is low enough performance (i.e. it’s an embedded chip, only 4TOPS) that the US has no intent to meddle.  It’s the 100+ TOPS chips that US is blocking. 

b. 不会, Myriad X 可以销售,性能稍低 (嵌入式芯片 4TOPS),被禁的是100+TOPS的芯片。


a. So we are keeping price as low as possible for the hardware.  For example the Azure Kinect DK only does depth and is $400.  We do depth, AI, video encoding, lossless zooming, feature tracking, and a ton of other things for $149 in KickStarter, $299 after KickStarter.

b. 我们尽可能的降低硬件售价。 比如Azure Kinect DK只有深度功能,但是要$400。 我们做深度、人工智能、图像处理、无损耗的缩放、物体的追踪 一系列功能,只需要$149 在Kickstarter上,活动结束后是$299。


a. Yes, it is available now for full MSRP.  $199 for OAK-1 and $299 for OAK-D.

b. 有的, 原价是$199 OAK-1, $299 OAK-D。


a. Speed and depth capability.  OpenMV is about 0.05 TOPS whereas OAK is 4.0 TOPS.  So for CV and AI tasks it is about 80 times faster.

b. 速度以及深度上有区别。 OpenMV只有0.05TOPS, 但是OAK有4.0TOPS,并且人工智能以及图像处理要比OpenMV快80倍。


a.  Yes, we have many tutorials for a variety of models.  Some examples are here: https://docs.luxonis.com/training/

b. 我们提供很多不同的教程,在这个链接上可以看https://docs.luxonis.com/training/


a. So the best sales location is likely any stores focusing on computer vision.

b. 任何图像处理相关的商店应该都可以购买。


a. It is Intel Movidius Myriad X.

b. 是Intel的Modvidius Myriad X.


a. ArduCam will likely carry this in China.

b. 在中国ArduCam应该会卖这个。

9、100多刀,上生产环境太贵了吧?还不如Jetson nano呢,对比一下有什么优势吗?

a. So Jetson Nano does not include a camera.  To buy the camera on this design alone it is $200.  It is a 12MP/60Hz very high-end camera.

b. Jetson Nano并不带摄像头, 单独购买这个摄像头就得$200。


a. It is Intel Movidius Myriad X, but works w/ ARM, Intel, AMD, any operating system that can run OpenCV.

b. 这是Intel的Movidius Myriad X, 但是可以用作任何ARM, Intel, AMD, 任何操作系统上 (只要能支持OpenCV)。


a. Yes, it is open source.  See here: https://github.com/luxonis/depthai-hardware

b. 是的开源,链接请看以上。


a. Yes, partnership.  So we were developing this as Open Source platform already and CEO of OpenCV reached out about doing this through OpenCV officially.  We thought was great idea so we’re doing it!  Been working together for a long time now.

b. 合作关系, 因为我们是开源 并且OpenCV的CEO 联系了我们 所以我们就一起合作了。


a. Can buy now here:  https://www.kickstarter.com/projects/opencv/opencv-ai-kit?ref=1fhhds

b. We are also working to get going with ArduCam to sell in China.

c. 可以在以上链接购买, 并且我们可能会让ArduCam 在国内进行销售。

14、Nvidia 是不是也可以?

a. Yes, this works on NVIDIA Tx1, Tx2, Jetson Nano, Xavier NX, etc. - all of them. 

b. 是的,可以在Nvidia 所有的板子上运行。

15、Intel 开发这个也是用在hybrid mode 吗???

a. So yes Intel definitely is involved closely. 

b. 是的Intel肯定是完全闭合的。


a. Yes, the API supports C++ as well instead of Python.  And microPython will be supported running directly on-chip

b. 是的,API支持C++ 和Python,并且microPython也将会支持。


a. Yes

b. 能


a. No

b. 不需要


a. Yes, please feel free to reach out to brandon@luxonis.com.  We can get you volume pricing.

b. 可以的, 请邮件brandon@luxonis.com 我们可以提供团购价。


a. Neural inference (e.g. Object detection, image classification of the ROI of a detected object, etc.) 神经网络计算, 物体识别, 图像分类

b. 3D object localization (both monocular object detection plus stereo depth and stereo neural inference supported) 3D空间定位, 深度预测

c. Object tracking 物体追踪

d. Stereo depth 深度预测

e. h.264/h.265 encoding h.264/h.265编码

f. Digital zoom/ePTZ (leveraging the full 12MP sensor resolution... which is 6x full 1080p streams) 缩放 并且有12MP的传感器像素 就是六倍的1080p摄像头

g. Background subtraction 背景处理

h. Feature tracking 特征追踪

i. Motion estimation 表情预测

j. Arbitrary crop/rescale/reformat and ROI return 图像处理(裁剪,缩放,格式改变)


a. Yes (可以)


a. Yes.  Already one in community is writing this.  There is already ROS as well.

b. 支持, 已经有位开发者在进行开发,并且ROS也会支持。

23、硬件的Schemtatic 会提供吗

a. Yes, here: https://github.com/luxonis/depthai-hardware

b. 是的, 请看以上链接。


a. Yes, here: https://github.com/luxonis/depthai  Many others here: https://docs.luxonis.com/faq/#githubs

b. 是的,请开以上链接。


a. Any CPU training platform will work.  Microsoft Azure, AWS, Google Colab, etc. or local computer.

b. 任何支持CPU训练平台都可以。 微软的Azure, 亚马逊的AWS, 谷歌的Colab, 或者本地电脑。

26、所以一般Ubuntu 即可嗎?

a. Yes, that’s right.  Ubuntu 16, 18, 20 all work, and really any Linux works.  And Mac OS X and Windows also work.

b. 可以的,Ubuntu系统16、18、20都可以, 任何Linux的都可以, 并且Mac OS X和 Window都可以。


a. We wrote additional libraries that leverage OpenCV and have open-sourced those.

b. 我们有增加一些库并且开源。

28、还是在intel 的Openvino平台对吧?主要用他们的推理框架吗?

a. So OpenVINO is used to optimize the models to run faster.

b. OpenVINO用于优化让模型运行更快些。

29、挺不錯的 前期要先安裝哪些套件 只要open champion 即可?才可以跑呢?

a. The dependencies are light.  And effectively the same as OpenCV.

b. 非常轻便, 就像OpenCV一样方便。


a. We are working on this.

b. 我们在运作这个。


a. Not yet. 

b. 目前不行。


a. Yes, it is here for now: https://shop.luxonis.com/collections/all/products/depthai-usb3c-with-onboard-cameras-mount/?aff=2

b. 是的,可以在以上链接购买。

33、Nvidia card 和你们有什么差异吗?

a. Yes so NVIDIA Jetson Nano is 0.472 TOPS and our solution is 4.0 TOPS - so about 8x faster total CV/AI processing capability.  And has super nice 12MP camera built in.  And is much lower power.

b. Nvidia 的Jetson Nano只有0.472 TOPS,但是我们有4.0TOPS, 所以是大约8倍的运行速度。 并且我们有嵌入式的12MP 摄像头。 消耗功率也更小。