Porting CaffeOnACL to RK3399 Android platform
The release is based on Rockchip RK3399 Platform
- ACL/NEON
- ACL/GPU(OpenCL)
- OpenBLAS
- Mixed Mode
Tested with Android NDK r15c and cmake 3.5.1 on Ubuntu 16.04
# Install some dependencies
sudo apt-get install python-dev python-pip git cmake
sudo pip install pyyaml six
# Download source code and prebuilts
git clone --recursive https://github.com/RockchipOpensourceCommunity/CaffeOnACL-Android.git
cd CaffeOnACL-Android
export NDK_ROOT=<path/to/your/ndk/root>
# Build Caffe
./build_caffe.sh
# Download network
./model_download.sh
# Prepare model data for RK3399 device
./model_prepare.sh
# Update binary data for RK3399 device
./binary_update.sh
# Running SqueezeNet network (default is AlexNet, support AlexNet, GoogLeNet, SqueezeNet, MobileNet)
./model_runner.py SqueezeNet
| Library | Version |
|---|---|
| Arm Compute Library | v17.10 |
| OpenBLAS | v0.2.20 |
| Boost | v1.65 |
| glog | v0.3.3 |
| gflags | v2.2.0 |
| protobuf | v3.1.0 |
| lmdb | v0.9.21 |
| OpenCV | v3.3.0 |