An end benchmark application for Android phones is made by computer vision researchers
An end benchmark application for Android phones is made by computer vision researchers - Bertemu lagi dengan saya Admin Red Line Techno, Kali ini saya menulis artikel dengan judul An end benchmark application for Android phones is made by computer vision researchers, semoga saja isi postingan mengenai
android tech news, bermanfaat buat sahabat Red Line Techno.
Judul : An end benchmark application for Android phones is made by computer vision researchers
link : An end benchmark application for Android phones is made by computer vision researchers
A team of computer researchers from Etis Zurich wants to improve the AII development in smartphones. Intuitive: They created a benchmark system to evaluate the performance of major neural network architectures used for normal AI tasks.
It is useful for other AI researchers, and they expect chip makers (through helping them get insight into insights). Android developers (see how many of their AI models look fast on their devices); By showing that you have the necessary drivers for the AI accelerator on a particular device, the phone is free to blows. (Therefore, whether or not they should believe the marketing messages of a company.)
The AP Benchmark application is available for download in the Google Play Store. It works with Android's 4.1 or more devices. Researchers have suggested that the AI's performance is "the ultimate destiny".
A definition benchmark system is image image, image recognition, image dee billing, image super resolution, photo enhancement or segmentation.
It is also at this point that some algorithms used in self-controlled driving systems are also checked. However, there is no practical need at this stage. Anyway so far. Researchers say it is not clear that using a hardware platform for automatic driving - if the device is too fast to be used for the future, "is likely to be at least as likely as possible.)
Help users evaluate the result and the current state of the artist gets an experience in the modern districts.
The researchers expect the DxOMark-like universal certified metric that is used to evaluate the camera performance - all benchmarks in benchmark are open source. Current ranks of various smartphones and mobile processors are available on the project's web page.
AI research and producer Andre Ignatov said the Benchmark system and application developed within three months.
The display score reflects two main points: SoC's speed and available RAM.
"Look at one of the two devices: one of a score of 6000, 200 of the score. First of all, a few fingers, five-stage device to pravarttikkumenkil, which means that it takes 30 times that of the second device. This means that approximately 2.5 minutes. We think this is spidinepparriyalla aplikkesanukalekkuriccan, such as face recognition, but rather the approach to Benefit Ksamatayekkuriccan: try to identify their phone to wait up to 10 seconds.
"The memory is about to be large: the network / input image is bigger - the processor needs more RAM to process, ie the phone has a smaller RAM, such as just 0.3MP photo, so this improvement is obviously unusable, but if it is possible for full HD images to do the same, Opportunities are opened, so the higher the score - the more Nkirnnamaya use algorithms to / be able to process large images / It will take some time. "
The benchmark concept suggests that "liabilities" in this lab are for research and industry, "the limitations for running a recent AI algorithm in smartphones have become curiosities at some point."
"This information (except that, all AI algorithms work remotely on the server, not on your device, but with a few built-in apps that have synchronized the phone's firmware), we have decided to develop our own device which clearly demonstrates the performance and capacity of each device," he adds.Ignatov is interesting because of the current state of the smartphones after the benchmarketing system and the scores on the Android devices.
For instance, the team found that the devices running the Qualcomm chips were not the winners they had won - ie, 845 AI capabilities of Snapdragon and 8x performance speeds were based on the promotional material of the company.
"This acceleration is only available for 'Qantas' networks and can not be deployed on the phone so you can not take for 'normal' networks," he said. "The most saddening fact is that they are theoretically the fastest chains for themselves, Not yet implemented. Now the only way to get this acceleration is Snapdragon Protector SDK. Only the processors. As a result, if you develop an application that uses the AI, Snapdragon SoCs will not be available anymore and only have been developed for their processors. "
The researchers found that Huawei's Kirin 970 CPU is much less than snapdragon in the technology - 636 - a surprisingly strong performance.
"Their integrated NPN provides nearly 10x acceleration for networks, so the most powerful phone CPIs and GPUs can not compete with it," Ignatov says. In addition, the Huawei P20 / P20 Pro is the only smartphone running on the Android 8.1 operating system. Currently AI acceleration is provided.
Huawei is not great news for the phone owners, but Ignato believes that the NPU is not capable of 'quantified' networks. (He also suggests that the company will assure that this support will be added by the end of this year); And it uses its own RAM - it's a "very limited" size, so "you can not process large images" ...
"If they solve these two problems - next year (with them) no one will be able to compete with them," he emphasizes, and emphasizes that this assessment merely refers to a socialism. The Huawei processor does not have NPU module.
For Samsung's processors, all the company's devices still operate on Android 8-10, but only Android Acceleration on Android 8.1 and above is available. See.
The performance of the CPU is "significantly different" - up to 50% on the same Samsung device - due to the thrilling and power optimization logic. Then Ai
"Megapetik" helps Chiplake to speed up both "fast" networks and "normal" networks - that is, the "high CPU" performance.
However, Ignatov called on the company "to lead the Edge-A Technology Revolution" by Ignatov - "something more than their dreams". "Do not refer to Kiran's 970 NPU with the hood without using any of the same proximity to the Samsung Xenox CP."
"In short: Snapdragon - theoretically good results, but it's lack of drivers; Huawei - the best results now in the nearest future; Samsung does not have any acceleration support now (this will change much soon after their own AI chip develops), but powerful CPUs; The good results for MediaTek - mid range devices, but of course Munnerramalla ayum. "
Furthermore, some of the results were samples from sample, not the spirit of the smit, so the team's website was not included in the benchmark table.
"Some changes will still be made and we will wait until the market comes with the final firmware," he added.
Terima kasih telah membaca artikel An end benchmark application for Android phones is made by computer vision researchers dengan alamat link https://redlinetechno.blogspot.com/2018/08/an-end-benchmark-application-for.html
Judul : An end benchmark application for Android phones is made by computer vision researchers
link : An end benchmark application for Android phones is made by computer vision researchers
It is useful for other AI researchers, and they expect chip makers (through helping them get insight into insights). Android developers (see how many of their AI models look fast on their devices); By showing that you have the necessary drivers for the AI accelerator on a particular device, the phone is free to blows. (Therefore, whether or not they should believe the marketing messages of a company.)
The AP Benchmark application is available for download in the Google Play Store. It works with Android's 4.1 or more devices. Researchers have suggested that the AI's performance is "the ultimate destiny".
A definition benchmark system is image image, image recognition, image dee billing, image super resolution, photo enhancement or segmentation.
It is also at this point that some algorithms used in self-controlled driving systems are also checked. However, there is no practical need at this stage. Anyway so far. Researchers say it is not clear that using a hardware platform for automatic driving - if the device is too fast to be used for the future, "is likely to be at least as likely as possible.)
Help users evaluate the result and the current state of the artist gets an experience in the modern districts.
The researchers expect the DxOMark-like universal certified metric that is used to evaluate the camera performance - all benchmarks in benchmark are open source. Current ranks of various smartphones and mobile processors are available on the project's web page.
AI research and producer Andre Ignatov said the Benchmark system and application developed within three months.
The display score reflects two main points: SoC's speed and available RAM.
"Look at one of the two devices: one of a score of 6000, 200 of the score. First of all, a few fingers, five-stage device to pravarttikkumenkil, which means that it takes 30 times that of the second device. This means that approximately 2.5 minutes. We think this is spidinepparriyalla aplikkesanukalekkuriccan, such as face recognition, but rather the approach to Benefit Ksamatayekkuriccan: try to identify their phone to wait up to 10 seconds.
"The memory is about to be large: the network / input image is bigger - the processor needs more RAM to process, ie the phone has a smaller RAM, such as just 0.3MP photo, so this improvement is obviously unusable, but if it is possible for full HD images to do the same, Opportunities are opened, so the higher the score - the more Nkirnnamaya use algorithms to / be able to process large images / It will take some time. "
The benchmark concept suggests that "liabilities" in this lab are for research and industry, "the limitations for running a recent AI algorithm in smartphones have become curiosities at some point."
"This information (except that, all AI algorithms work remotely on the server, not on your device, but with a few built-in apps that have synchronized the phone's firmware), we have decided to develop our own device which clearly demonstrates the performance and capacity of each device," he adds.Ignatov is interesting because of the current state of the smartphones after the benchmarketing system and the scores on the Android devices.
For instance, the team found that the devices running the Qualcomm chips were not the winners they had won - ie, 845 AI capabilities of Snapdragon and 8x performance speeds were based on the promotional material of the company.
"This acceleration is only available for 'Qantas' networks and can not be deployed on the phone so you can not take for 'normal' networks," he said. "The most saddening fact is that they are theoretically the fastest chains for themselves, Not yet implemented. Now the only way to get this acceleration is Snapdragon Protector SDK. Only the processors. As a result, if you develop an application that uses the AI, Snapdragon SoCs will not be available anymore and only have been developed for their processors. "
The researchers found that Huawei's Kirin 970 CPU is much less than snapdragon in the technology - 636 - a surprisingly strong performance.
"Their integrated NPN provides nearly 10x acceleration for networks, so the most powerful phone CPIs and GPUs can not compete with it," Ignatov says. In addition, the Huawei P20 / P20 Pro is the only smartphone running on the Android 8.1 operating system. Currently AI acceleration is provided.
Huawei is not great news for the phone owners, but Ignato believes that the NPU is not capable of 'quantified' networks. (He also suggests that the company will assure that this support will be added by the end of this year); And it uses its own RAM - it's a "very limited" size, so "you can not process large images" ...
"If they solve these two problems - next year (with them) no one will be able to compete with them," he emphasizes, and emphasizes that this assessment merely refers to a socialism. The Huawei processor does not have NPU module.
For Samsung's processors, all the company's devices still operate on Android 8-10, but only Android Acceleration on Android 8.1 and above is available. See.
The performance of the CPU is "significantly different" - up to 50% on the same Samsung device - due to the thrilling and power optimization logic. Then Ai
"Megapetik" helps Chiplake to speed up both "fast" networks and "normal" networks - that is, the "high CPU" performance.
However, Ignatov called on the company "to lead the Edge-A Technology Revolution" by Ignatov - "something more than their dreams". "Do not refer to Kiran's 970 NPU with the hood without using any of the same proximity to the Samsung Xenox CP."
"In short: Snapdragon - theoretically good results, but it's lack of drivers; Huawei - the best results now in the nearest future; Samsung does not have any acceleration support now (this will change much soon after their own AI chip develops), but powerful CPUs; The good results for MediaTek - mid range devices, but of course Munnerramalla ayum. "
Furthermore, some of the results were samples from sample, not the spirit of the smit, so the team's website was not included in the benchmark table.
"Some changes will still be made and we will wait until the market comes with the final firmware," he added.
Bagaimana postingan Artikel tentang An end benchmark application for Android phones is made by computer vision researchers ?
Mungkin cukup sekian postingan An end benchmark application for Android phones is made by computer vision researchers kali ini, semoga bisa memberi manfaat untuk anda. Sampai bertemu kembali di postingan artikel selanjutnya. Jangan lupa bagikan artikel ini jika bermanfaat.
Terima kasih telah membaca artikel An end benchmark application for Android phones is made by computer vision researchers dengan alamat link https://redlinetechno.blogspot.com/2018/08/an-end-benchmark-application-for.html
0 Response to "An end benchmark application for Android phones is made by computer vision researchers"
Post a Comment