GitPlanet

Cheap and reliable Node.js hosting starts at $3/month, and $1/month static HTML hosting

Created with love in Canada, visit hostnodejs.com today

Feel like to post an Ad? Learn Details
All Projects → freedomtan → glDelegateBench

freedomtan / glDelegateBench

Licence: BSD-3-Clause license
quick and dirty inference time benchmark for TFLite gles delegate

Programming Languages

java
68154 projects - #9 most used programming language

Labels

benchmarkgpucompute-shadertensorflow-litetflitetflite-delegatetflite-gpu-delegate

Projects that are alternatives of or similar to glDelegateBench

glDelegateBenchmark
quick and dirty benchmark for TFLite gles delegate on iOS
Stars: ✭ 13 (-23.53%)
Mutual labels:  benchmark, tensorflow-lite, tflite, tflite-delegate, tflite-gpu-delegate
Selfie2Anime-with-TFLite
How to create Selfie2Anime from tflite model to Android.
Stars: ✭ 70 (+311.76%)
Mutual labels:  tensorflow-lite, tflite
android tflite
GPU Accelerated TensorFlow Lite applications on Android NDK. Higher accuracy face detection, Age and gender estimation, Human pose estimation, Artistic style transfer
Stars: ✭ 105 (+517.65%)
Mutual labels:  tensorflow-lite, tflite
E2E-tfKeras-TFLite-Android
End to end training MNIST image classifier with tf.Keras, convert to TFLite and deploy to Android
Stars: ✭ 17 (+0%)
Mutual labels:  tensorflow-lite, tflite
TFLite-Mobile-Generic-Object-Localizer
Python TFLite scripts for detecting objects of any class in an image without knowing their label.
Stars: ✭ 42 (+147.06%)
Mutual labels:  tensorflow-lite, tflite
Tensorflow-lite-kotlin-samples
📌This repo contains the kotlin implementation of TensorflowLite Example Android Apps🚀
Stars: ✭ 17 (+0%)
Mutual labels:  tensorflow-lite, tflite
tflite-vx-delegate
Tensorflow Lite external delegate based on TIM-VX
Stars: ✭ 28 (+64.71%)
Mutual labels:  tflite, tflite-delegate
DiscEval
Discourse Based Evaluation of Language Understanding
Stars: ✭ 18 (+5.88%)
Mutual labels:  benchmark
php-framework-benchmark
php framework benchmark (include laravel、symfony、silex、lumen、slim、yii2、tastphp etc)
Stars: ✭ 17 (+0%)
Mutual labels:  benchmark
StreamBench
Measuring the performance of popular streaming engines with Yahoo's Streaming Benchmark
Stars: ✭ 52 (+205.88%)
Mutual labels:  benchmark
Drishti
Drishti is an open-source cross-platform mobile application project at Incubate Nepal that incorporates Machine Learning and Artificial Intelligence to help visually impaired people recognize different currency bills and perform daily cash transactions more effectively. We plan to expand Drishti to other applications like Short Text and Document…
Stars: ✭ 23 (+35.29%)
Mutual labels:  tflite
Shuhai
Shuhai is a benchmarking-memory tool that allows FPGA programmers to demystify all the underlying details of memories, e.g., HBM and DDR4, on a Xilinx FPGA
Stars: ✭ 53 (+211.76%)
Mutual labels:  benchmark
nodemark
A modern benchmarking library for Node.js
Stars: ✭ 23 (+35.29%)
Mutual labels:  benchmark
OptimisationAlgorithms
Searching global optima with firefly algorithm and solving traveling salesmen problem with genetic algorithm
Stars: ✭ 20 (+17.65%)
Mutual labels:  benchmark
perforator
Record "perf" performance metrics for individual functions/regions of an ELF binary.
Stars: ✭ 33 (+94.12%)
Mutual labels:  benchmark
nntrainer
NNtrainer is Software Framework for Training Neural Network Models on Devices.
Stars: ✭ 92 (+441.18%)
Mutual labels:  tensorflow-lite
gobench
A benchmark framework based on Golang
Stars: ✭ 50 (+194.12%)
Mutual labels:  benchmark
surface nets
Pet project to learn compute shaders.
Stars: ✭ 35 (+105.88%)
Mutual labels:  compute-shader
DeepLearningBenchmarks
Benchmarks across Deep Learning Frameworks in Julia and Python
Stars: ✭ 24 (+41.18%)
Mutual labels:  benchmark
micro-runner
Micro-Runner, a CLI playground for benchmarking your JavaScript code
Stars: ✭ 27 (+58.82%)
Mutual labels:  benchmark
View All Similar Projects ➔

glDelegateBench

quick and dirty inference time benchmark for TFLite gles delegate

The TensorFlow team announced TFLite GPU delegate and published related docs [2][3] in Jan 2019. But except Mobilenet V1 classifier, there is no publicly available app to evaluate it, so I wrote a quick and dirty app to evaluate other models.

For the 4 public models mentioned in [1], I got the following numbers on Pixel 2.

model name CPU 1 thread (ms) CPU 4 threads (ms) GPU (ms)
Mobilenet 150 75 21
PoseNet 183 96 40
DeepLab V3 219 131 91
Mobilenet SSD V2 COCO 264 158 49

On Xiaomi Mi 9, I got

model name CPU 1 thread (ms) CPU 4 threads (ms) GPU (ms)
Mobilenet 39 35 15
PoseNet 48 47 19
DeepLab V3 61 64 65
Mobilenet SSD V2 COCO 69 75 36

On Pixel 3a, I got

model name CPU 1 thread (ms) CPU 4 threads (ms) GPU (ms)
Mobilenet 113 80 52
PoseNet 138 96 78
DeepLab V3 173 132 144
Mobilenet SSD V2 COCO 200 167 113

Check https://github.com/freedomtan/glDelegateBenchmark/ for iOS code

add a local_tflite_aar branch to test ruy, the new TFLite CPU backend

on Pixel 2, I got

model name CPU 1 thread (ms) CPU 4 threads (ms) GPU (ms)
Mobilenet 117 37 20
PoseNet 140 47 39
DeepLab V3 177 72 122
Mobilenet SSD V2 COCO 202 75 60

on Pixel 3a, I got

model name CPU 1 thread (ms) CPU 4 threads (ms) GPU (ms)
Mobilenet 107 44 51
PoseNet 131 57 77
DeepLab V3 164 82 145
Mobilenet SSD V2 COCO 184 86 113

Update Oct 31, 2019. Nightly aar binaries are with ruy and OpenCL backend

Update Dec 8, 2019, Dec for Pixel 3a came with DSP and GPU NNAPI 1.2 driver, so we can have NNAPI numbers on Pixel 3a

on Pixel 2 (w/ libOpenCL-pixel.so from Pixel 3), I got

model name CPU 1 thread (ms) CPU 4 threads (ms) GPU OpenCL (ms) GPU GL Compute Shader (ms)
Mobilenet 118 34 10 21
PoseNet 142 43 14 41
DeepLab V3 174 75 21 69
Mobilenet SSD V2 COCO 202 73 18 48

on Pixel 3a, I got

model name CPU 1 thread (ms) CPU 4 threads (ms) GPU (ms) NNPAI (ms)
Mobilenet 107 44 28 25
PoseNet 131 57 38 32
DeepLab V3 164 82 60 186
Mobilenet SSD V2 COCO 184 86 54 249

on Pixel 4, I got

model name CPU 1 thread (ms) CPU 4 threads (ms) GPU Delegate (ms) NNAPI (ms)
Mobilenet 42 13 8 7
PoseNet 52 15 11 11
DeepLab V3 66 25 20 98
Mobilenet SSD V2 COCO 70 24 16 86

[1] https://medium.com/tensorflow/tensorflow-lite-now-faster-with-mobile-gpus-developer-preview-e15797e6dee7

[2] https://www.tensorflow.org/lite/performance/gpu

[3] https://www.tensorflow.org/lite/performance/gpu_advanced

Note that the project description data, including the texts, logos, images, and/or trademarks, for each open source project belongs to its rightful owner. If you wish to add or remove any projects, please contact us at [email protected].