Drown_Recognition

1. Prepare

Make sure clone all the file inside this github page.
Some importance data、model、video put inside the GoogleDrive.
Put all the files inside the same path of the directories.

1.1 Check the environment

Devices:

1. Ubuntu 16.04 LTS
2. Memory: 15.6 GiB
3. Processor: Intel® Core™ i5-9400F CPU @ 2.90GHz × 6
4. Graphics : GeForce RTX 2080/PCIe/SSE2
5. Os Type: 64-bit

Dependency:

1. Cuda compilation tools, release 10.0, V10.0.130
2. Cudnn : 7.4.2
3. Python : 3.6.12
4. tensorflow-gpu: 1.15.0
5. keras : 2.3.1

Check Cuda Version:

nvcc -V

Check Cuda Version:

cat /usr/local/cuda/include/cudnn.h | grep CUDNN_MAJOR -A 2
or
cat /usr/include/cudnn.h | grep CUDNN_MAJOR -A 2

1.2 Create new environment

Make sure already install Anaconda.

if not:

1. Download the { }.sh file. here
2. Install Anaconda:

bash ./{filename.sh}

Completed install Anacconda:

1. Create new environment and activate:

conda create -n {env_name} python=3.6.12
conda activate {env_name}

1.3 Install Tensorflow GPU version

1. Install tensorflow-gpu 1.15.0 version(make sure your tensorflow-gpu version is fit to your cuda version)

pip3 install tensorflow-gpu==1.15.0

2. Check if your tensorflow-gpu completed install or not.

python
import tensorflow as tf
tf.__version__

1.15.0

2. Install OpenPose

2.1 Install dependencies(Tensorflow version)

cd src/depends/tf-pose-estimation
pip3 install -r requirements.txt

2.2 Download Pretrained models(MobileNet-Thin models)

MobileNet-Thin models has been inside, see folder:

src/depends/tf-pose-estimation/models/graph
cmu  mobilenet_thin  mobilenet_v2_large  mobilenet_v2_small

Choose to download(CMU models)

Original Version more accuracy but large and long execution time:

cd models/graph/cmu
bash download.sh

2.3 Install Libraries

cd ../../../
pip3 install jupyter tqdm
pip3 install keras==2.3.1
sudo apt install swig

2.4 Compile c++ library

pip3 install "git+https://github.com/philferriere/cocoapi.git#egg=pycocotools&subdirectory=PythonAPI"
cd src/depends/tf-pose-estimation/tf_pose/pafprocess
swig -python -c++ pafprocess.i && python3 setup.py build_ext --inplace

2.5 final install some library

cd ../../../../../
pip3 install -r requirements.txt

2.6 Final Check~

Test on video:

python src/s5_test.py \
    --model_path model/trained_classifier.pickle \
    --data_type video \
    --data_path data_test/exercise.avi \
    --output_folder output

if error occurred!

cd src/depends/tf-pose-estimation
python setup.py install

Then try Final check again.

3. NOW TRAINING!

Reference:

1. Chenge Yang, Zhicheng Yu, Feiyu Chen, "Human Pose Estimation Benchmarking and Action Recognition," Deep Learning Project, Winter 2019, Northwestern University, 26-Oct-2019.
2. Feiyu Chen, "Human-Pose-Estimation-Benchmarking-and-Action-Recognition,"Deep Learning Project, Winter 2019, Northwestern University, 26-Oct-2019.
3. jiajunhua, "ildoonet-tf-pose-estimation," 10-Apr-2019
4. philferriere ,"cocoapi,"25-Oct-2018
5. 橘白卯咪,"欸你在幹嘛?跟一眼瞬間就懂你的行為辨識技術初次見面," 01-Sep-2020
6. Zhe Cao, Gines Hidalgo, Tomas Simon, Shih-En Wei, Yaser Sheikh,"OpenPose: Realtime Multi-Person 2D Pose Estimation using Part Affinity Fields,"Computer Vision and Pattern Recognition,18-Dec-2018