Action Recognition Modeule(FSA-CNN) using 2D skeleton extracted fromm ETRI-Activity3D dataset.
The accuracy is 91.00% for Testset.
- Python = 3.6.8
- Tensorflow-gpu or tensorflow = 1.12.0
- Keras = 2.2.4
.
├── TestBed_OpenPose_v4_COCO_6_9100.h5 # Weight file
├── Test_Code.py # Test code that consists of reading samples, loading models with trained weights and testing
├── Training_Code.py # Training code using ETRI-Activity3D Dataset
├── LICENSE.md
├── LICENSE_ko.md
└──README.md
-
clone this git
-
Download ETRI-Activity3D_Mat file from
https://drive.google.com/drive/folders/1KrLsDfJS9nfTZwBB52TEBVT3nHN2yY8e?usp=sharing
and unzip at the root folder.
(every .mat files should be at just inside of "ETRI-Activity3D_Mat" folder)
- install the requirements:
pip install keras==2.2.4
pip install tensorflow-gpu==1.12.0
pip install libpython or conda install libpython
(maybe random, math, numpy and os modules are included in libpython)
run
"python Training_Code.py"
If you want to set pre-trained weights as initialization,
Unlock the comment of line 211("network.load_weights(weight_path)").
You can get "Weight_save_temp.h5" as weights of the latest epoch, and
"Weight_save.h5" as weights of the best test accuracy during your training.
run
"python Test_Code.py"
If you want to initialize using your own weights,
change line 31("weight_path = 'TestBed_OpenPose_v4_COCO_6_9100.h5' ")
to your weight file.
This software is a part of AIR, and follows the AIR License and Service Agreement.
Jang, J., Kim, D., Park, C., Jang, M., Lee, J., & Kim, J. (2020). ETRI-Activity3D: A Large-Scale RGB-D Dataset for Robots to Recognize Daily Activities of the Elderly. arXiv preprint arXiv:2003.01920.