Anton Lokhmotov

I've provided descriptions for the following image classification images: - TFLite: - [image-classification-tflite.ubuntu-18.04](https://hub.docker.com/r/ctuning/image-classification-tflite.ubuntu-18.04) - [image-classification-tflite.ubuntu-16.04](https://hub.docker.com/r/ctuning/image-classification-tflite.ubuntu-16.04) - [image-classification-tflite.centos-7](https://hub.docker.com/r/ctuning/image-classification-tflite.centos-7) - [image-classification-tflite.debian-9](https://hub.docker.com/r/ctuning/image-classification-tflite.debian-9) - ArmNN-TFLite: - [image-classification-armnn-tflite.debian-9](https://hub.docker.com/r/ctuning/image-classification-armnn-tflite.debian-9) - TF-C++: - [image-classification-tf-cpp.debian-9](https://hub.docker.com/r/ctuning/image-classification-tf-cpp.debian-9)

The following object-detection images need to be pushed: ``` $ docker system df -v | grep object-detection ctuning/object-detection-armnn-tflite.debian-9 latest 8587785e3298 8 minutes ago 2.331GB 1.839GB 492.6MB 0 ctuning/object-detection-tflite.debian-9 latest bdf725180c00...

I've also created a cool "dashboard" Docker image to benchmark the host machine on `docker build` and then display the results interactively similar to http://cknowledge.org/dashboard/mlperf.mobilenets  On the screenshot above,...

As far as I know, TFLite only provides GPU acceleration via [AndroidNN](https://developer.android.com/ndk/guides/neuralnetworks/), which is available from Android 8.1. Unfortunately, the latest phones we have only support Android 8.0. If someone...

> Also the problem is that all frameworks don't share at least one model, so I can't compare them directly. Now you can! Please take a look at our brand...

> ArmCL 18.05 OpenCL: MobileNets v1 0.25 128 (Looks strange that it have size of 141 Mb) The model itself is only ~2 MB but we bundle together the engine...

> In my benchmarks TFLite CPU faster then ArmCL(for MobileNets v1 0.25 128) and Caffe CPU faster then Caffe OpenCL(for SqueezeNet 1.1) That's expected for very small models. There's simply...

While HiKey960 is a development board, it has the same chip (Hisilicon Kirin960) that Huawei used in their several popular phones (including Mate 9 Pro and P10). I have results...

I've added TFLite results on Huawei Mate 10 Pro (HiSilicon Kirin 970) and Samsung Galaxy S8 US (Qualcomm Snapdragon 835). You may want to filter the results by `Library=tflite-0.1.7`, `Version=1`...

> it will be great if anyone can share link with current 'view' of dashboard Thanks for your feedback! Yes, supporting links with settings is on our roadmap. > Also...