# -*- coding: utf-8 -*-
import glob
import dlib
import cv2
import pandas as pd
import datetime
import os

#候補画像を格納したディレクトリパス
IMG_PATH = "./data/NNC-Challenge_Dataset/*.jpg"

def face_detect(img_list):
    # print(img_list)
    # 顔領域を検出する識別器呼び出し
    detector = dlib.get_frontal_face_detector()
    image_class_data = pd.DataFrame(index=[])
    img_count = 1

    for img_path in img_list:
        img = cv2.imread(img_path, cv2.IMREAD_COLOR)
        cv_img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
        faces = detector(cv_img, 1)
        # 顔領域が検出出来た場合
        if faces:
            image_class_data = image_class_data.append([[img_path,"0","there are people in pic(maybe..)"]])
            for face in faces:
                # 候補画像サイズ取得
                height, width = img.shape[:2]
                # 顔領域の座標点取得
                top = face.top()
                bottom = face.bottom()
                left = face.left()
                right = face.right()
                # イレギュラーな顔領域は無視
                if top < 0 or left < 0 or bottom > height or right > width:
                    print("イレギュラー",top,bottom,left,right)
                    image_class_data = image_class_data.append([[img_path,"1","ID"+str(img_count)+": face detect is not good shape(maybe..)"]])
                    DIR_split = img_path.split("/")[-1]
                    os.system("cp "+img_path+" ./classification_faceonly/1/"+str(img_count) + '.jpg')
                    break
                # 顔領域をトリミング
                dst_img = img[top:bottom, left:right]
                # 顔画像サイズを正規化して保存
                face_img = cv2.resize(dst_img, (64,64))
                new_img_name = '././classification_faceonly/0/'+str(img_count) + '.jpg'
                cv2.imwrite(new_img_name, face_img)
                img_count += 1
        # 顔領域が検出出来なかった場合
        else:
            DIR_split = img_path.split("/")[-1]
            os.system("cp "+img_path+" ./classification_faceonly/2/"+DIR_split)
            image_class_data = image_class_data.append([[img_path,"2","no people(maybe..)"]])

        # dev用
        # if img_count > 60:
        #     image_class_data.columns = ["directory","classfication","reason"]
        #     image_class_data.to_csv("./out_facedetect_dev_"+str(datetime.datetime.today())+".csv",index=False)

    image_class_data.columns = ["directory","classfication","reason"]
    image_class_data.to_csv("./out_facedetect_"+str(datetime.datetime.today())+".csv",index=False)

if __name__ == '__main__':
    images = glob.glob(IMG_PATH)  
    face_detect(images)

# -*- coding: utf-8 -*-
# trainとvalidationに分けたcsvファイルを生成

import numpy as np
import pandas as pd
import glob 
import datetime

image_class_data0 = pd.DataFrame(index=[],columns=["x:image","y"])
image_class_data = pd.DataFrame(index=[],columns=["x:image","y"])

l0 = glob.glob("./classification_faceonly/0/anotation/0/*.jpg")
l1 = glob.glob("./classification_faceonly/0/anotation/1/*.jpg")

# カメラ目線と非カメラ目線の画像の枚数を同じにする
leng_0 = len(l0)
leng_1 = len(l1)

if leng_1 > leng_0:
    leng_1 = leng_0
else:
    leng_0 = leng_1

# y: 0 (カメラ目線) 
image_class_data0["x:image"] = l0[:leng_0]
num_array = np.zeros(leng_0)
image_class_data0["y"] = num_array

# y: 1 (非カメラ目線)
image_class_data["x:image"] = l1[:leng_1]
num_array = np.ones(leng_1)
image_class_data["y"] = num_array

print(image_class_data0.head())
print(image_class_data0.describe())
print(image_class_data.head())
print(image_class_data.describe())

# trainとvalidationに分ける (8:2)
image_class_data0_train = image_class_data0[:int(leng_0*0.8)]
image_class_data0_validation = image_class_data0[int(leng_0*0.8):]
image_class_data_train = image_class_data[:int(leng_0*0.8)]
image_class_data_validation = image_class_data[int(leng_0*0.8):]

# trainとvalidationそれぞれでconcat処理
df_concat_train = pd.concat([image_class_data0_train, image_class_data_train])
df_concat_validation = pd.concat([image_class_data0_validation, image_class_data_validation])

print(df_concat_train.describe())
print(df_concat_validation.describe())

df_concat_train.to_csv("./dataset_train.csv",index=False)
df_concat_validation.to_csv("./dataset_validation.csv",index=False)