search

gWOLF3 / rscnn

A cnn framework on Android platform, implemented with RenderScript and java, support MobileNet-SSD and faster-rcnn. No JNI/C++ or thirdparty dependencies.
MIT License
0 stars 1 forks source link

readme

Render Script CNN for Android

This repo is a fork of chuanqi305's rscnn, a CNN framework based on RenderScript.

The purpose of this repo is to demonstrate an implementation for deploying prebuilt detection models from Caffe (using MobileNetv1 + SSD) onto an Android device.

What You'll Need

1) Android Studio w/ API 28

2) Caffe (w/ssd) : https://github.com/chuanqi305/ssd

3) Anaconda (Recomended)

Usage

  1. Download MobileNet-SSD model. 
    git clone https://github.com/chuanqi305/MobileNet-SSD

You'll want to download the pretrained MobileNetSSD_deploy.caffemodel and move it to MobileNet-SSD folder.

  1. Use script/convert_caffe_model.py to convert the model to new format, do not forget to change the caffe root path in the converting script.
python script/convert_caffe_model.py --model MobileNet-SSD/deploy.prototxt --weights MobileNet-SSD/MobileNetSSD_deploy.caffemodel --savedir mobilenet-ssd
  1. Push the converted model files to src/main/assets folder of your project. 
    cp -ar mobilenet-ssd demo/src/main/assets/
  2. Run this demo, and you can select a photo to see the object detection result.

Errors

See Wiki for more documentation details of additional errors you might run into and how to correct them.