When I first came into contact with the horned sunnym, I was awkward. I even said that it is difficult to understand why there is a good artificial intelligent visual suite to take such a "sword walking" name, do not understand Chinese corner hummingbird. The deep meaning of the deep meaning, more of the HORNED SUNGEM of English, except for difficulties, there seems to be no feature. However, after a very chance, I went deep into the company's developer's developer's appearance. After the company, I had a great interest in this product. After getting the hives, it has been studying how it is studying. What can I do?
It may make you a shock, and now Herch has a name behind the intelral Intel Movidius NCS, there is also a shadow of this company. Before Movidius is not acquired by Intel, it has been tightly cooperated with Movidius. I even said that in addition to Movidus Irs yourself, the touch is unlimited is what I know most about MYRIAD 2, how to play strength, is it expected? Let's take a look at the next hive Ai visual kit.
Corner Hummingbird Ai Vision Kit
From the packaging, the definition of the touch of the poker is very clear. The definition of the visual suite is very clear, "Plug and Ai", this can be more clear than the name of the suite itself, it is very intuitive, which means plugging so that you can enjoy the collision with AI.
There are not many internal accessories, even if the root data line is not provided, you need to purchase extra. Note that it is the data line of the interface of USB TPYE-C. It may be careful netizens will ask, why is using USB TYPE-C data line? The meaning of behind is a surprise, because the angle humming bird Ai visual kit supports USB3.0, of course, is also compatible with USB2.0.
The "small card" included in the package is nothing information, but there is a website above, the content on the website includes the basic use of the AI visual suite, some of the neural network models included, some examples, and angle hummingbirds API and other content.
The bee hummingbird is an instant AI visual kit, which consists of a small motherboard based on Movidius Myriad 2 VPU + a camera module. If you have played Intel Movidius NCS, congratulations, the talents, the honeybird this kit. There is basically no difficulty, or even you can use it as a combination of NCS + cameras. Therefore, in the article in 2018 domestic engineer / Chuangke's most worthy of the development board TOP 10 rankings were rated as a representative of blue and blue.
It is clear that the MOVIDIUS Myriad 2 VPU is clear, and the heat of this VPU is really not covered, so heat dissipation is also done on the angle humbra kit.
Corner Hummingbird Ai Vision Kit Disassembly
When removing this as a comparative LOW's Yarik shell holder, the main board of the hive vision suite is still surprised.
The FPC connection of the CSI interface is used between the camera module and the motherboard. It seems that the FPC size is customized. The two-headed PIN foot size is different, such as the camera like the Raspberry Pie does not necessarily be used directly.
The motherboard is really small, almost adult thumbs like the size, but the board is not exquisite, it may be the first batch, some of the components are somewhat awkward, but it does not affect the use.
The chamfer is the CSI interface in the green box box for connecting the outside of the camera;
The two-side black box is the reserved binocular camera interface, which can achieve stereo visual system. Of course, this part has not yet been developed, and the ability to do it yourself.
The red frame is RCOH's RT5C619 power management unit;
The yellow part is Movidius Ma2450 VPU, it is necessary to mention that there are many segmental segments in artificial intelligence, and MA2450 is mainly for image and video reasoning acceleration. You can look at the Myriad 2 SOC internal architecture.
MYRIAD 2 integrates 12 128-Bit Shave Vector Processors, and many hardware accelerator, image / vision signal processors, two 32-bit RISC processors, which can effectively accelerate video and images, Myriad 2 SOC Some of the interfaces support are as follows:
12 Lanes Mipi, each LANE supports 1.5Gbps data bandwidth and can be configured as CSI-2 or DSI
Support I2C, SPI, I2S, PWM, USB3.0 (integrated PHY), SDIO, Ethernet (1Gbit)
The specific performance indicators of Myriad 2 can handle data from 12 13 megapixel cameras at the same time of the frame rate of 48fps, that is, take 4K video with 60FPS is uncomfortable, some GPUs on the market, Myriad 2 The main advantage is to maintain lower power consumption while reaching this performance, such as using a USB interface to power.
Of course, you may have to 怼 Let me, low power consumption? The advantages of four words, edge computing, and low power consumption are the applications that need to have stronger image / video processing capabilities on the current market, such as service-type robots, drones, AR / VR and other devices. MYRIAD 2 has a very big advantage.
The USB TYPE-C interface in the yellow box in the figure below provides power and transfer data for the board;
In the red frame, it is an internal debug interface, not open.
The green box is the 64Mbit SPI Flash GD25LQ64 of Zhaoyi Innovation;
In the Blue box, the Pericom Pi3USB3102 USB3.0 & USB2.0 switch allows the angle of hives ai visual kit to simultaneously support USB2.0 or USB3.0 HOST hosts.
In general, the angle hummingbird Ai visual suite provides an image-based edge computing device for your products through the integrated front section, and you can work with Ubuntu, Macos, Raspbian, or even Android. DIY smart camera, smart robots, etc. Unique AI applications, which are specific to the actual experience.
Used
Here is a UP Squared Board with a Ubuntu system.
To use neural network models and routines that look at the show, you need to install the hives development tools they provide, enter directly in the Command Window:
Git clone https://github.com/hornedsungem/sungemsdk.git
Then wait for the download to be completed.
Installing SDK, you can see SDK-free SDK in instager's files can support MacOS, Raspbian, and Ubuntn systems, which can meet users of different platforms.
After the installation is complete, you can find 7 ready-made routines in the Examples / Python directory: hello2018.py, facedtector.py, objectdector.py, sketchguess.py, getImage.py, ImageRecongNition.py, SceneRecorder.py.
Take Hello2018.py as an example, of course, this is not a simple Hello routine. In fact, this application to the mnist digital identification model, you can look at the code that is unlimited, it is really simple, and it is nothing wrong with you usually write Python syntax. And the most critical model, if you don't use it yourself, you can use it. Of course, because of the relationship between model training data, it may not be accurate enough, and it is necessary to subdivision, or you need your own training model.
This routine is the picture in the identification / MISC / 2018_Mnist / directory, a total of four, as shown below
The result of reasoning output, Hello 2018, is completely correct.
In addition, I also tried all of the other routines, but because of the training data, the accuracy rate is relatively low, the accuracy is relatively high, and the accuracy is relatively universal "object", such as people, such as people. Chair, banana, etc..
The identification of a picture below is more interesting, and people are identified, but the human hand is constructed by the keyboard. It is indeed more like a chair. The result is that it is true to identify the chair. It can be seen that the road to machine is still very long.
When I identify some alternatives, the accuracy is relatively low, such as:
As another example:
In addition, in the actual test, it is found that sometimes the program is loaded into the new model, and it has been prompted to swap the USB line. According to the unlimited engineer of the touch, this problem belongs to the software small bug, which has been solved in the test, after which I will update SDK. .
Overall, although the beekestophils have some small bugs on software, these issues are completely solved by later optimization, nor the big problem, and the unlimited angle bird Ai visual suite is designed by this unique front-end integration. At the same time, simultaneous implementation of the actual field is achieved into direct data such as object categories, face comparison results, and for the development of such applications, this is an extremely convenient development kit. Of course, because of the relationship of training data, the model comes with the angle hummingbird itself is not very ideal. I want to really apply the developer on my own product, but also need my own neural network to travel, I believe that the scene is unlimited. Demo just better helps developers learn how to use the APIs they provide, so don't drill the horn, and go.
summary
Artificial intelligence has developed to this year, the fact that the artificial intelligence we have to face again today should be a whole system, if only rely on algorithms, hardware is not good, the big data support for training is extremely important, so, When some relatively forward edge calculate the product, don't deny its innovation due to the bug on the surface, the angle hummingbird Ai visual suite is a suite of the blue and the blue, such as it, even supports the Android system. It may still have a lot of problems, but do not affect the use, and extremely convenient "Plug and Ai" plug-and-play and unique front-end integrated design, perhaps open the artificially intelligent DIY boom.
By the way, I still have to spit a sentence, I don't deliberately remember, I will never have a honeybird in the next!
Corner Hummingbird Artificial Intelligent Vision Kit
Price ¥ 699.00
A artificial intelligent visual kit based on the concept of "Plug and Ai" is committed to providing public, using and creates AI applications, defining the opportunity of AI life.
Ai Getting Started Learning Kit Special Popular Products
Go to buy
you may also like:
2018 Domestic Engineer / Chuangguan is the most worthy of development board TOP 10 ranking
Our other product:
