Skip to content

Latest commit

 

History

History
122 lines (99 loc) · 3.94 KB

README.md

File metadata and controls

122 lines (99 loc) · 3.94 KB
Raw

Extended Android Pose Estimation using Mace(Xiaomi Inference Computing Engine)

pose_bts_jimin pose_jumping_jacks

Inference Results (Model size About 8MB (.a + .bin) PCKh : About 83)

Mate 10 About 40 ms
A2 Lite About 170 ms
Xiaomi Mi pad About 40 ms
Device Inferene time
Galaxy S8+ About 60 ms
Galaxy S8 About 70 msms
Galaxy S7 About 100 ms(require to more testing)
Galaxy S6 110 ms

Inference method

  • Compute the more accurate x and y coordinate by weight averaging 3x3 Slide window array around it without using the max x, y coordinate.
  • Before weight averaging coordinate x, y 33, 48 after 33.23634, 48.47124 (If you reverse it to the display size of the phone, the difference becomes more bigger.)

This repository is implemented PoseEstimationForMobile (Thanks for authors). Model files were excluded(related to company). If you want to write a model file, please use the PoseEstimationForMobile Model. or frozen_model.pb --> Optimizing model --> Optimized_model.pb --> Converting Deployment model --> Finally Make model.a file.**

if you use this repository, convert model(s) to code
Example)
# model_graph_format: code
# model_data_format: code
builds
  ├── include
  │   └── mace
  │       └── public
  │           ├── mace_engine_factory.h
  │           └── cpm_v2.h
  └── model
      └── cpm-v2.a

PoseEstimationForMobile' repository is used v0.9.0 (2018-07-20) So migration lastest version v0.10.0 (2019-01-03). If you build a Mace library (libmace.a) from Xiaomi github (master or v0.10.0 later version), you need to re-write the code(*.cpp). and deleted OpenCV3 library

v0.10.0 (2019-01-03) highlighting the change portion.

Improvements
  1. Support mixing usage of CPU and GPU.
  2. Support ONNX format.
  3. Support ARM Linux development board.
  4. Support CPU quantization.
  5. Update DSP library.
  6. Add Depthwise Deconvolution of Caffe.
  7. Add documents about debug and benchmark.
  8. Bug fixed.
Incompatible Changes
  1. Remove all APIs in mace_runtime.h
New APIs
  1. Add GPUContext and GPUContextBuilder API.
  2. Add MaceEngineConfig API.
  3. Add MaceStatus API.
  4. MaceTensor support data format.
etc..
My project environment
  • OS Platform and Distribution: macOS Mojave 10.14
  • Android SDK version : 21 later (Camera2 API)
  • Android NDK version : 16(If you higher NDK version(17, 18), you replace cmake arguments gnustl_static => c++_shared or c++_static. because higher version is not supported a gnustl_static)
  • Xiaomi version: v0.10.0
  • TensorFlow version: master branch
  • Python version: 3.6
  • Bazel version (if compiling from source): 18 and 19
  • GCC/Compiler version (if compiling from source): 5.4
Advanced model's option
  • target_abis : armeabi-v7a
  • model_graph_format : code
  • model_data_format : code
  • winograd : 4 (But there is no performance's improvement)