predict.py

from __future__ import absolute_import, division, print_function, unicode_literals
from keras.preprocessing import sequence
from keras.datasets import imdb
from keras import layers, models
from keras.models import Sequential
from keras import layers
import os
import sys
import pickle
import numpy as np
from tensorflow.keras.preprocessing.text import Tokenizer
from tensorflow.keras.utils import to_categorical
import random
from keras import optimizers
from keras.layers import SimpleRNN, Dense
from keras.layers import Bidirectional
import tensorflow as tf
from numpy import argmax
import argparse

def load_data(dirname):
    if dirname[-1]!='/':
        dirname=dirname+'/'
    listfile=os.listdir(dirname)
    X = []
    Y = []
    for file in listfile:
        if "_" in file:
            continue
        wordname=file
        textlist=os.listdir(dirname+wordname)
        for text in textlist:
            if "DS_" in text:
                continue
            textname=dirname+wordname+"/"+text
            numbers=[]
            #print(textname)
            with open(textname, mode = 'r') as t:
                numbers = [float(num) for num in t.read().split()]
                #print(len(numbers[0]))
                for i in range(len(numbers),4200):
                    numbers.extend([0.000])
            landmark_frame=[]
            row=0
            for i in range(0,35):
                landmark_frame.extend(numbers[row:row+84])
                row += 84
            landmark_frame=np.array(landmark_frame)
            landmark_frame=landmark_frame.reshape(-1,84)
            X.append(np.array(landmark_frame))
            Y.append(wordname)
    X=np.array(X)
    Y=np.array(Y)
    print(Y)
    x_train = X
    x_train=np.array(x_train)
    return x_train,Y


#prediction: lấy từng label trong file label.txt
def load_label():
    listfile=[]
    with open("label.txt",mode='r') as l:
        listfile=[i for i in l.read().split()]
    label = {}  #khởi tạo 1 dict
    count = 1
    for l in listfile:
        if "_" in l:
            continue
        label[l] = count
        count += 1
    return label
    
def main(input_data_path,output_data_path,data_path):
    comp='bazel build -c opt --define MEDIAPIPE_DISABLE_GPU=1 \mediapipe/examples/desktop/multi_hand_tracking:multi_hand_tracking_cpu'
    cmd='GLOG_logtostderr=1 bazel-bin/mediapipe/examples/desktop/multi_hand_tracking/multi_hand_tracking_cpu \
  --calculator_graph_config_file=mediapipe/graphs/hand_tracking/multi_hand_tracking_desktop_live.pbtxt'
    listfile=os.listdir(input_data_path)
    if not(os.path.isdir(output_data_path+"Relative/")):
        os.mkdir(output_data_path+"Relative/")
    if not(os.path.isdir(output_data_path+"Absolute/")):
        os.mkdir(output_data_path+"Absolute/")
    for file in listfile:
        if not(os.path.isdir(input_data_path+file)): #ignore .DS_Store
            continue
        word = file+"/"
        fullfilename=os.listdir(input_data_path+word)
        if not(os.path.isdir(output_data_path+"_"+word)):
            os.mkdir(output_data_path+"_"+word)
        if not(os.path.isdir(output_data_path+"Relative/"+word)):
            os.mkdir(output_data_path+"Relative/"+word)
        if not(os.path.isdir(output_data_path+"Absolute/"+word)):
            os.mkdir(output_data_path+"Absolute/"+word)
        os.system(comp)
        #outputfilelist=os.listdir(output_data_path+'_'+word)
        for mp4list in fullfilename:
            if ".DS_Store" in mp4list:
                continue         
            inputfilen='   --input_video_path='+input_data_path+word+mp4list
            outputfilen='   --output_video_path='+output_data_path+'_'+word+mp4list
            cmdret=cmd+inputfilen+outputfilen
            os.system(cmdret)

    #output_dir=output_data_path
    x_test,Y=load_data(data_path)
    new_model = tf.keras.models.load_model('model.h5')
    #new_model.summary()

    labels=load_label()
    print(labels)

    xhat = x_test
    yhat = new_model.predict(xhat)
    print(yhat[0])
    #print(yhat[1])
    #print(yhat[2])
    predictions = np.array([np.argmax(pred) for pred in yhat])
    rev_labels = dict(zip(list(labels.values()), list(labels.keys())))
    s=0
    txtpath=output_data_path+"result.txt" 
    with open(txtpath, "w") as f:
        for i in predictions:
            f.write("true_label: ")
            f.write(Y[s])
            f.write(" === ")
            f.write("predict_label: ")
            f.write(rev_labels[i])
            f.write("\n")
            s+=1

if __name__ == "__main__":
    parser = argparse.ArgumentParser(description='Predict Sign language with Mediapipe')
    parser.add_argument("--input_data_path",help=" ")
    parser.add_argument("--output_data_path",help=" ")
    parser.add_argument("--data_path",help=" ")
    args=parser.parse_args()
    input_data_path=args.input_data_path
    output_data_path=args.output_data_path
    data_path=args.data_path
    main(input_data_path,output_data_path,data_path)