A Simple But High-accuracy LSTM for human Action Recognition

• matlab_m/: transform the given dataset(NTU RGB+D) to your need , from ".skeleton" to ".mat"

  • demo.m: an example for you to transform the dataset using the given functions .You shall verify the "fileFolder","dirOutput",and "savepath"
  • classfile.m: divide the "*.mat" file according to their class.
  • read_skeleton_file.m: a function to read the skeleton files (given by NTU RGB dataset)
  • savetomat.m: a function to save the skeleton data from skeleton files to mat files
  • show_skeleton_on_depthmaps.m: a function to show the skeleton information on the depthmaps(thanks to the NTU RGB+D dataset)
  • show_skeleton_on_IR_frames.m: a function to show the skeleton information on the IR frames(thanks to the NTU RGB+D dataset)
  • show_skeleton_on_RGB_frames.m: a function to show the skeleton information on the RGB frames(thanks to the NTU RGB+D dataset)

• lstm_py/: the train and test python file using tensorflow lib.

  • main.py: the train python file using tensorflow.
  • evaluate.py: the test file to evaluate your model perfermance.
  • mtop.py: transform the skeleton files form ".mat" to ".npy" for python files . Also, you may use it for seperate train and test set .
  • model_lstm/: well-trained model of lstm .

• keras: the train and test python file using keras lib.

  • main.py: an example for you to transform the dataset using the given functions .You shall verify the "train_file", and "test_file"

    Requirements

• code only tested on linux system (ubuntu 16.04)

• Python 3 (Anaconda 3.6.3 specifically) with numpy and matplotlib

• Tensorflow

• keras

• matlab

  model structure

  

  To prepare using the given data by NTU RGB+D

  Using matlab (from ".skeleton" to ".mat")

  In file demo.m

  fileFolder=['D:\research\ntuRGB\ske_f\',num2str(t),'\'];%using your own dataset path
  savepath=['D:\research\ntuRGB\mat_f\',num2str(t),'\'];%using your own save path

  In file classfile.m

  SOURCE_PATH_t =[ 'D:\research\ntuRGB\mat_f\',num2str(i),'\'];%using your own "*.mat" files path  
  DST_PATH_t1 = [ 'D:\research\ntuRGB\mat_f\',num2str(i),'\test'];%using your own wanted test set saved path
  DST_PATH_t2 = [ 'D:\research\ntuRGB\mat_f\',num2str(i),'\train'];%using your own wanted train set saved path

  matlab demo.m
  matlab classfile.m

  Using python (from ".mat" to ".npy")

  In file mtop.py

  trainpath='./CS/train/'#verify your train data files forder here ("*.mat" file)
  testpath='./CS/test/'#verify your test data files forder here ("*.mat" file)

  Using Tensorflow

  To train

  In file main.py

  train_file='CV_20/train' #verify your train data files forder here
   test_file='CV_20/test' #verify your train data files forder here
   model_file="model/my-model.meta"#verify your train model data file
  model_path="model/"#verify your train model data folder

python lstm_py/main.py

• you will get your own model saved in the "model/"

  To test

  In file evaluate.py

  train_file='CV_20/train' #verify your train data files forder here
   test_file='CV_20/test' #verify your train data files forder here
   model_file="model/my-model.meta"#verify your train model data file
  model_path="model/"#verify your train model data folder

python lstm_py/evaluate.py

In file main.py

train_file='CV_20/train' #verify your train data files forder here 
test_file='CV_20/test' #verify your train data files forder here 
model_file="model/my-model.meta"#verify your train model data file
model_path="model/"#verify your train model data folder

python keras/main.py