HFNet_VINO

KO | EN

gitlite — search

#ai #golang #android #machine-learning #deep-learning #linux #docker #java #windows #cli #awesome-list #pytorch #security #ios

(★ 12)

Using OpenVINO to accelerate HF-Net

HFNet_VINO Latest Version Download

Download Latest Version (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Accelerate HF-Net using Openvino

In this project, we adapted the source code of hfnet for openvino tensorflow support that can be successfully convert to openvino model. After converted to openvino, the fps is increased from 6 to 22.

The original code is [Here](https://github.com/ethz-asl/hfnet)

# How to:

1. Donwload the trained weights from [here](https://drive.google.com/drive/folders/1B2jSg_H5BSXjNq8iAFis1aQ4vt2FAv-y?usp=sharing) and convert the graph for inference.

```
cd hfnet/hfnet
python3 export_model.py ../../hfnet_ckpt/config.yaml ../../ --exper_name ../../hfnet_ckpt
```
And then you will get a folder `saved_model` which contain the whole inference graph and weights.

2. Convert the `saved_model` to openvino format.

* You needs to install the master branch of [OpenVINO](https://github.com/opencv/dldt)

```
python3 opt/intel/openvino/deployment_tools/model_optimizer/mo.py --saved_model_dir ./saved_models
```

and then you will get the `saved_model.bin` and `saved_model.xml`.

if you are using OpenVINO 2019R3, you may failed to convert the model, here is the [solution](https://github.com/openvinotoolkit/openvino/issues/335).

**here is my converted model [xml](https://drive.google.com/file/d/1dVTk8AURVvH8fBsMkziSNYqIcUIWvcV_/view?usp=sharing), [bin](https://drive.google.com/file/d/1DPp4j4p3ytxZGZGtTMQUE1AF-ZbLL4q_/view?usp=sharing)**

3. testing the model:

after done all steps before, you can run ```python3 tf_camera_demo.py``` for using tensorflow get the HFNet outputs or run ```python3 vino_camera_demo.py``` for using Openvino get the HFNet ouputs.

Feel free to open any issue if you need to help:)

# ROS Usage

The major problem with ROS is that ROS does not support Python3 officially. Though it is possible to import rospy in Python3, cv_bridge must be recompiled locally.

```
# install some tools for building
sudo apt-get install python-catkin-tools python3-dev python3-numpy

# make a new catkin workspace
mkdir ~/cv_bridge_ws && cd ~/cv_bridge_ws

# config and build -- check the following path and revise them if needed (e.g. if you use Python 3.6 or higher)
catkin config -DPYTHON_EXECUTABLE=/usr/bin/python3 -DPYTHON_INCLUDE_DIR=/usr/include/python3.5m -DPYTHON_LIBRARY=/usr/lib/x86_64-linux-gnu/libpython3.5m.so
catkin config --install
mkdir src; cd src
git clone -b melodic https://github.com/ros-perception/vision_opencv.git
cd ..
catkin build cv_bridge
```

If having the error of `Could NOT find Boost (missing: python3)`, edit src/vision_opencv/cv_bridge/CMakeLists.txt and change `Boost REQUIRED python3` to `Boost REQUIRED python3.5` (or any other minor versions).

Before running the script, set up PYTHONPATH properly by
```
. /opt/ros/kinetic/setup.bash
. ~/cv_bridge_ws/install/setup.bash --extend
. /opt/intel/openvino/setupvars.sh
```

And run the demo:
```
python3 ros_demo.py
```

![DEMO](https://raw.githubusercontent.com/ahuizxc/HFNet_VINO/master/demo_vino.gif)

HFNet_VINO

HFNet_VINO Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

HFNet_VINO

HFNet_VINO Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

Download README