## Author: SHIVANK SHARMA
## Modified: Toshio Iwata
## Released under Apache 2.0 license 

import numpy as np
import pandas as pd
import os
import cv2
import matplotlib.pyplot as plt
import tensorflow as tf
from tensorflow import keras
from PIL import Image
from sklearn.model_selection import train_test_split
from tensorflow.keras.preprocessing.image import ImageDataGenerator
from tensorflow.keras.optimizers import Adam
from sklearn.metrics import accuracy_score
np.random.seed(42)

from matplotlib import style
style.use('fivethirtyeight')

data_dir = './'
train_path = './train_1'
test_path = './'

# Resizing the images to 30x30x3
IMG_HEIGHT = 30
IMG_WIDTH = 30
channels = 3

NUM_CATEGORIES = len(os.listdir(train_path))
NUM_CATEGORIES

# Label Overview
### Modified for lisa
classes = { 0:'bike',
            1:'car', 
            2:'human', 
            3:'white', 
            4:'beju', 
            5:'black', 
            6:'wh2bk', 
            7:'bk2wh', 
            8:'wh_bk', 
            9:'bk_wh' }

###################################################
## for making models
###################################################
# モデルを作成
model1 = keras.Sequential(name="my_model1")
# model1を確認
print(model1)
model1.add(keras.layers.Conv2D(filters=16, kernel_size=(3,3), activation='relu', input_shape=(IMG_HEIGHT,IMG_WIDTH,channels)))


model2 = keras.Sequential(name="my_model2")
# model2を確認
print(model2)
model2.add(keras.layers.Conv2D(filters=32, kernel_size=(3,3), activation='relu', input_shape=(28,28,16)))
model2.add(keras.layers.MaxPool2D(pool_size=(2, 2)))


model3 = keras.Sequential(name="my_model3")
# model3を確認
print(model3)
model3.add(keras.layers.Conv2D(filters=64, kernel_size=(3,3), activation='relu', input_shape=(13,13,32)))


model4 = keras.Sequential(name="my_model4")
# model4を確認
print(model4)
model4.add(keras.layers.Conv2D(filters=128, kernel_size=(3,3), activation='relu', input_shape=(11,11,64)))
model4.add(keras.layers.MaxPool2D(pool_size=(2, 2)))

model4B = keras.Sequential(name="my_model4B")
model4B.add(keras.layers.Flatten(input_shape=(4,4,128)))


model5 = keras.Sequential(name="my_model5")
# model5を確認
print(model5)
model5.add(keras.layers.Dense(128, activation='relu', input_dim=2048))
model5.add(keras.layers.Dropout(rate=0.5))

model5B = keras.Sequential(name="my_model5B")
model5B.add(keras.layers.Dropout(rate=0.5, input_dim=128))

model6 = keras.Sequential(name="my_model6")
# model6を確認
print(model6)
###model6.add(keras.layers.Dense(43, activation='softmax', input_dim=128))
### Modified for lisa
model6.add(keras.layers.Dense(10, input_dim=128)) # 2023/03/08

import csv

###################################################
## for read
###################################################
########################### lay1
lay1 = model1.layers[0]
tmpw = lay1.get_weights()

with open('w1_int.csv', mode='r') as f:
    reader = csv.reader(f)
    r = 0
    c = 0
    i = 0
    k = 0
    print("w1 shape : ", lay1.get_weights()[0].shape)
    for row in reader:
#        print(row)
        for k in range(0, lay1.get_weights()[0].shape[3]):
            tmpw[0][r][c][i][k] = row[k] ### [0]縺ｫ驥阪∩菫よ焚
        c += 1
        if(c == range(lay1.get_weights()[0].shape[1]).stop):
            c = 0
            r += 1
            if(r == range(lay1.get_weights()[0].shape[0]).stop):
                r = 0
                i += 1

with open('b1_int.csv', mode='r') as f:
    reader = csv.reader(f)
    print("b1 shape : ", lay1.get_weights()[1].shape)
    for row in reader:
#        print(row)
        for k in range(0, lay1.get_weights()[1].shape[0]):
            tmpw[1][k] = row[k] ### [1]縺ｫ繝舌う繧｢繧ｹ

lay1.set_weights(tmpw)
print("w1 read done")

########################### lay2
lay2 = model2.layers[1-1]
tmpw = lay2.get_weights()

with open('w2_int.csv', mode='r') as f:
    reader = csv.reader(f)
    r = 0
    c = 0
    i = 0
    k = 0
    print("weight shape : ", lay2.get_weights()[0].shape)
    for row in reader:
#        print(row)
        for k in range(0, lay2.get_weights()[0].shape[3]):
            tmpw[0][r][c][i][k] = row[k] ### [0]縺ｫ驥阪∩菫よ焚
        c += 1
        if(c == range(lay2.get_weights()[0].shape[1]).stop):
            c = 0
            r += 1
            if(r == range(lay2.get_weights()[0].shape[0]).stop):
                r = 0
                i += 1

with open('b2_int.csv', mode='r') as f:
    reader = csv.reader(f)
    print("weight shape : ", lay2.get_weights()[1].shape)
    for row in reader:
#        print(row)
        for k in range(0, lay2.get_weights()[1].shape[0]):
            tmpf = float(row[k]) / 8 
            tmpi = int(tmpf)
            tmpw[1][k] = tmpi ### row[k] ### [1]縺ｫ繝舌う繧｢繧ｹ
            ###tmpw[1][k] = row[k] ### [1]縺ｫ繝舌う繧｢繧ｹ

lay2.set_weights(tmpw)
print("w2 read done")

########################### lay3
lay3 = model3.layers[3-3]
tmpw = lay3.get_weights()

with open('w3_int.csv', mode='r') as f:
    reader = csv.reader(f)
    r = 0
    c = 0
    i = 0
    k = 0
    print("weight shape : ", lay3.get_weights()[0].shape)
    for row in reader:
#        print(row)
        for k in range(0, lay3.get_weights()[0].shape[3]):
            tmpw[0][r][c][i][k] = row[k] ### [0]縺ｫ驥阪∩菫よ焚
        c += 1
        if(c == range(lay3.get_weights()[0].shape[1]).stop):
            c = 0
            r += 1
            if(r == range(lay3.get_weights()[0].shape[0]).stop):
                r = 0
                i += 1

with open('b3_int.csv', mode='r') as f:
    reader = csv.reader(f)
    print("weight shape : ", lay3.get_weights()[1].shape)
    for row in reader:
#        print(row)
        for k in range(0, lay3.get_weights()[1].shape[0]):
            tmpf = float(row[k]) / 8 
            tmpi = int(tmpf)
            tmpw[1][k] = tmpi ### row[k] ### [1]縺ｫ繝舌う繧｢繧ｹ
            ###tmpw[1][k] = row[k] ### [1]縺ｫ繝舌う繧｢繧ｹ

lay3.set_weights(tmpw)
print("w3 read done")

########################### lay4
lay4 = model4.layers[4-4]
tmpw = lay4.get_weights()

with open('w4_int.csv', mode='r') as f:
    reader = csv.reader(f)
    r = 0
    c = 0
    i = 0
    k = 0
    print("weight shape : ", lay4.get_weights()[0].shape)
    for row in reader:
#        print(row)
        for k in range(0, lay4.get_weights()[0].shape[3]):
            tmpw[0][r][c][i][k] = row[k] ### [0]縺ｫ驥阪∩菫よ焚
        c += 1
        if(c == range(lay4.get_weights()[0].shape[1]).stop):
            c = 0
            r += 1
            if(r == range(lay4.get_weights()[0].shape[0]).stop):
                r = 0
                i += 1

with open('b4_int.csv', mode='r') as f:
    reader = csv.reader(f)
    print("weight shape : ", lay4.get_weights()[1].shape)
    for row in reader:
#        print(row)
        for k in range(0, lay4.get_weights()[1].shape[0]):
            tmpf = float(row[k]) / 8 
            tmpi = int(tmpf)
            tmpw[1][k] = tmpi ### row[k] ### [1]縺ｫ繝舌う繧｢繧ｹ
            ###tmpw[1][k] = row[k] ### [1]縺ｫ繝舌う繧｢繧ｹ

lay4.set_weights(tmpw)
print("w4 read done")

########################### lay5
lay5 = model5.layers[7-7]
tmpw = lay5.get_weights()

with open('w5_int.csv', mode='r') as f:
    reader = csv.reader(f)
    i = 0
    k = 0
    print("weight shape : ", lay5.get_weights()[0].shape)
    for row in reader:
#        print(row)
        for k in range(0, lay5.get_weights()[0].shape[1]):
            tmpw[0][i][k] = row[k] ### [0]縺ｫ驥阪∩菫よ焚
        i += 1

with open('b5_int.csv', mode='r') as f:
    reader = csv.reader(f)
    print("weight shape : ", lay5.get_weights()[1].shape)
    for row in reader:
#        print(row)
        for k in range(0, lay5.get_weights()[1].shape[0]):
            tmpf = float(row[k]) / 8 
            tmpi = int(tmpf)
            tmpw[1][k] = tmpi ### row[k] ### [1]縺ｫ繝舌う繧｢繧ｹ
            ###tmpw[1][k] = row[k] ### [1]縺ｫ繝舌う繧｢繧ｹ

lay5.set_weights(tmpw)
print("w5 read done")

########################### lay6
lay6 = model6.layers[9-9]
tmpw = lay6.get_weights()

with open('w6_int.csv', mode='r') as f:
    reader = csv.reader(f)
    i = 0
    k = 0
    print("weight shape : ", lay6.get_weights()[0].shape)
    for row in reader:
#        print(row)
        for k in range(0, lay6.get_weights()[0].shape[1]):
            tmpw[0][i][k] = row[k] ### [0]縺ｫ驥阪∩菫よ焚
        i += 1

with open('b6_int.csv', mode='r') as f:
    reader = csv.reader(f)
    print("weight shape : ", lay6.get_weights()[1].shape)
    for row in reader:
#        print(row)
        for k in range(0, lay6.get_weights()[1].shape[0]):
            tmpf = float(row[k]) / 8 
            tmpi = int(tmpf)
            tmpw[1][k] = tmpi ### row[k] ### [1]縺ｫ繝舌う繧｢繧ｹ
            ###tmpw[1][k] = row[k] ### [1]縺ｫ繝舌う繧｢繧ｹ

lay6.set_weights(tmpw)
print("w6 read done")


test = pd.read_csv(data_dir + '/TestMovObj.csv') ### Modified for lisa

## Modified for Shiroume
def resize (image, size):
    h, w = image.shape[:2]
    aspect = w / h
    nh = nw = size
    if 1 >= aspect:
        nw = round(nh * aspect)
    else:
        nh = round(nw / aspect)
    resized = cv2.resize(image, dsize=(nw, nh))
    return resized

## Modified for Shiroume
def shiroume (image, size):
    resized = resize(image, size)

    h, w = resized.shape[:2]
    x = y = 0
    if h < w:
        y = (size - h) // 2
    else:
        x = (size - w) // 2
    resized = Image.fromarray(resized) #一旦PIL形式に変換
    canvas = Image.new(resized.mode, (size, size), (255, 255, 255)) # 黒埋めなら(0,0,0)にする
    canvas.paste(resized, (x, y))
    dst = np.array(canvas) #numpy(OpenCV)形式に戻す
    return dst

labels = test["ClassId"].values
imgs = test["Path"].values

data =[]

for img in imgs:
    try:
        image = cv2.imread(data_dir + '/' +img)

        ## Modified for Shiroume
        h, w = image.shape[:2]
        if h < w :
            val = w
        else :
            val = h
        ume_image = shiroume(image, val) #リサイズしたいサイズを指定

        image_fromarray = Image.fromarray(ume_image, 'RGB')
        resize_image = image_fromarray.resize((IMG_HEIGHT, IMG_WIDTH))
        data.append(np.array(resize_image))
    except:
        print("Error in " + img)
X_test = np.array(data)
####X_test = X_test/255

for n in range(0, X_test.shape[0]):
    for r in range(0, X_test.shape[1]):
        for c in range(0, X_test.shape[2]):
            tmpval = X_test[n][r][c][0] ### [0]にblue value
            tmpval = tmpval / 8
            X_test[n][r][c][0] = int(tmpval) * 8

for n in range(0, X_test.shape[0]):
    for r in range(0, X_test.shape[1]):
        for c in range(0, X_test.shape[2]):
            tmpval = X_test[n][r][c][1] ### [1]にgreen value
            tmpval = tmpval / 4
            X_test[n][r][c][1] = int(tmpval) * 4

for n in range(0, X_test.shape[0]):
    for r in range(0, X_test.shape[1]):
        for c in range(0, X_test.shape[2]):
            tmpval = X_test[n][r][c][2] ### [2]にred value
            tmpval = tmpval / 8
            X_test[n][r][c][2] = int(tmpval) * 8

X_test = np.asarray(X_test, dtype=int)

print("x_test shape : ", X_test.shape)
TNN = 0
#print(X_test[TNN])

##################################################### 
### write input image to a csv
##################################################### 
with open("Bin1_int.csv", mode='w') as f:
    for r in range(0, X_test.shape[1]):
        for c in range(0, X_test.shape[2]):
###            X_test[TNN][r][c][0] = 0
            tmpval = X_test[TNN][r][c][0] ### [0]にblue value
            f.write(str(tmpval))
            f.write(', ')
        f.write('\n')

with open("Gin1_int.csv", mode='w') as f:
    for r in range(0, X_test.shape[1]):
        for c in range(0, X_test.shape[2]):
###            X_test[TNN][r][c][1] = 255
            tmpval = X_test[TNN][r][c][1] ### [1]にgreen value
            f.write(str(tmpval))
            f.write(', ')
        f.write('\n')

with open("Rin1_int.csv", mode='w') as f:
    for r in range(0, X_test.shape[1]):
        for c in range(0, X_test.shape[2]):
###            X_test[TNN][r][c][2] = 0
            tmpval = X_test[TNN][r][c][2] ### [2]にred value
            f.write(str(tmpval))
            f.write(', ')
        f.write('\n')

### Modified for lisa
##################################################### 
outstr = "RGB565.csv"
with open(outstr, mode='w') as f:
    for i in range(20): ### 2023/09/12, use rndval and 20 pics
##    outstr = "RGB565_30x30_" + "{0:02d}".format(outn) + ".csv"
        if i%10 == 0 :
            rndval = int(i/10)
        elif i%10 == 1 :
            rndval = 180 + int(i/10)
        elif i%10 == 2 :
            rndval = 470 + int(i/10)
        elif i%10 == 3 :
            rndval = 1470 + 0 + int(i/10)
        elif i%10 == 4 :
            rndval = 1470 + 3 + int(i/10)
        elif i%10 == 5 :
            rndval = 1470 + 6 + int(i/10)
        elif i%10 == 6 :
            rndval = 1470 + 9 + int(i/10)
        elif i%10 == 7 :
            rndval = 1470 + 12 + int(i/10)
        elif i%10 == 8 :
            rndval = 1470 + 15 + int(i/10)
        elif i%10 == 9 :
            rndval = 1470 + 18 + int(i/10)
        for r in range(0, X_test.shape[1]):
            for c in range(0, X_test.shape[2]):
                tmpr = X_test[TNN+rndval][r][c][2] ### [2]にred value
                tmpr = tmpr >> 3 ### lower 3bit is 0
                tmpr = tmpr << 11
                tmpg = X_test[TNN+rndval][r][c][1] ### [1]にgreen value
                tmpg = tmpg >> 2 ### lower 2bit is 0
                tmpg = tmpg << 5
                tmpb = X_test[TNN+rndval][r][c][0] ### [0]にblue value
                tmpb = tmpb >> 3 ### lower 3bit is 0
                tmpval = tmpr + tmpg + tmpb
                f.write(str(tmpval))
                f.write(', ')
        f.write('\n')


##################################################### 
### start prediction
##################################################### 
lay1out = model1.predict(X_test) 
lay1out = lay1out / 256
lay1out = lay1out / 8
lay1out = np.asarray(lay1out, dtype=int)
lay1out = np.where(lay1out > 31, 31, lay1out)

##print(lay1out[TNN])

######### write lay1 output to a csv
print("lay1out shape : ", lay1out.shape)
outn = 0
for outn in range(lay1out.shape[3]):
    outstr = "lay1out_" + "{0:02d}".format(outn) + ".csv"
    with open(outstr, mode='w') as f:
        for r in range(0, lay1out.shape[1]):
            for c in range(0, lay1out.shape[2]):
                tmpval = lay1out[TNN][r][c][outn] ### channel outn
                f.write(str(tmpval))
                f.write(',')
            f.write('\n')


lay2out = model2.predict(lay1out) 
print("lay2out shape : ", lay2out.shape)
lay2out = lay2out / 256
lay2out = np.asarray(lay2out, dtype=int)
lay2out = np.where(lay2out > 31, 31, lay2out)
######### write lay2 output to a csv
print("lay2out shape : ", lay2out.shape)
outn = 0
for outn in range(lay2out.shape[3]):
    outstr = "lay2out_" + "{0:02d}".format(outn) + ".csv"
    with open(outstr, mode='w') as f:
        for r in range(0, lay2out.shape[1]):
            for c in range(0, lay2out.shape[2]):
                tmpval = lay2out[TNN][r][c][outn] ### channel outn
                f.write(str(tmpval))
                f.write(',')
            f.write('\n')


lay3out = model3.predict(lay2out) 
print("lay3out shape : ", lay3out.shape)
lay3out = lay3out / 256
lay3out = np.asarray(lay3out, dtype=int)
lay3out = np.where(lay3out > 31, 31, lay3out)
######### write lay3 output to a csv
print("lay3out shape : ", lay3out.shape)
outn = 0
for outn in range(lay3out.shape[3]):
    outstr = "lay3out_" + "{0:02d}".format(outn) + ".csv"
    with open(outstr, mode='w') as f:
        for r in range(0, lay3out.shape[1]):
            for c in range(0, lay3out.shape[2]):
                tmpval = lay3out[TNN][r][c][outn] ### channel outn
                f.write(str(tmpval))
                f.write(',')
            f.write('\n')


lay4out = model4.predict(lay3out) 
print("lay4out shape : ", lay4out.shape)
lay4out = lay4out / 256
lay4out = np.asarray(lay4out, dtype=int)
lay4out = np.where(lay4out > 31, 31, lay4out)
######### write lay4 output to a csv
print("lay4out shape : ", lay4out.shape)
outn = 0
for outn in range(lay4out.shape[3]):
    outstr = "lay4out_" + "{0:03d}".format(outn) + ".csv"
    with open(outstr, mode='w') as f:
        for r in range(0, lay4out.shape[1]):
            for c in range(0, lay4out.shape[2]):
                tmpval = lay4out[TNN][r][c][outn] ### channel outn
                f.write(str(tmpval))
                f.write(',')
            f.write('\n')

lay4Bout = model4B.predict(lay4out) 
print("lay4Bout shape : ", lay4Bout.shape)
lay4Bout = np.asarray(lay4Bout, dtype=int)
lay4Bout = np.where(lay4Bout > 31, 31, lay4Bout)
######### write lay4B output to a csv
print("lay4Bout shape : ", lay4Bout.shape)
outn = 0
outstr = "lay4Bout.csv"
with open(outstr, mode='w') as f:
    for r in range(0, lay4Bout.shape[1]):
        tmpval = lay4Bout[TNN][r]
        f.write(str(tmpval))
        f.write(',')


lay5out = model5.predict(lay4Bout) 
print("lay5out shape : ", lay5out.shape)
lay5out = lay5out / 256
lay5out = np.asarray(lay5out, dtype=int)
lay5out = np.where(lay5out > 31, 31, lay5out)

######### write lay5 output to a csv
print("lay5out shape : ", lay5out.shape)
outn = 0
outstr = "lay5out.csv"
with open(outstr, mode='w') as f:
    for r in range(0, lay5out.shape[1]):
        tmpval = lay5out[TNN][r]
        f.write(str(tmpval))
        f.write(',')

lay5Bout = model5B.predict(lay5out) 
print("lay5Bout shape : ", lay5Bout.shape)
lay5Bout = np.asarray(lay5Bout, dtype=int)
lay5Bout = np.where(lay5Bout > 31, 31, lay5Bout)

######### write lay5B output to a csv
print("lay5Bout shape : ", lay5Bout.shape)
outn = 0
outstr = "lay5Bout.csv"
with open(outstr, mode='w') as f:
    for r in range(0, lay5Bout.shape[1]):
        tmpval = lay5Bout[TNN][r]
        f.write(str(tmpval))
        f.write(',')

lay6out = model6.predict(lay5Bout) 
###lay6out = lay6out / 256 # 2023/03/08
print("lay6out shape : ", lay6out.shape)
pred = np.argmax(lay6out, axis=-1) 
lay6out = np.asarray(lay6out, dtype=int)
lay6out = lay6out >> 6 

### Modified for lisa
######### write lay6 output to a csv
print("lay6out shape : ", lay6out.shape)
for i in range(20): ### 2023/07/06, result of lay6 for 20 pictures
    if i%10 == 0 :
        rndval = int(i/10)
    elif i%10 == 1 :
        rndval = 180 + int(i/10)
    elif i%10 == 2 :
        rndval = 470 + int(i/10)
    elif i%10 == 3 :
        rndval = 1470 + 0 + int(i/10)
    elif i%10 == 4 :
        rndval = 1470 + 3 + int(i/10)
    elif i%10 == 5 :
        rndval = 1470 + 6 + int(i/10)
    elif i%10 == 6 :
        rndval = 1470 + 9 + int(i/10)
    elif i%10 == 7 :
        rndval = 1470 + 12 + int(i/10)
    elif i%10 == 8 :
        rndval = 1470 + 15 + int(i/10)
    elif i%10 == 9 :
        rndval = 1470 + 18 + int(i/10)
    outstr = "lay6out_sample_" + "{0:02d}".format(i) + ".csv"
    ### outstr = "lay6out.csv"
    with open(outstr, mode='w') as f:
        for r in range(0, lay6out.shape[1]):
            tmpval = lay6out[TNN+rndval][r]
            f.write(str(tmpval))
            f.write(',') 

#Accuracy with the test data
print('Test Data accuracy: ',accuracy_score(labels, pred)*100)
'''
errno = 0
plt.figure(figsize = (25, 25))
for i in range(1479):
    prediction = pred[i]
    actual = labels[i]
    if prediction != actual:
        print('prediction error ', i, ' : pred ', pred[i], ' : actual ', labels[i])
        plt.subplot(3, 6, errno + 1)
        plt.grid(False)
        plt.xticks([])
        plt.yticks([])
        plt.imshow(X_test[i])
        errno = errno + 1
'''

from sklearn.metrics import confusion_matrix
cf = confusion_matrix(labels, pred)

import seaborn as sns
df_cm = pd.DataFrame(cf, index = classes,  columns = classes)
plt.figure(figsize = (20,20))
sns.heatmap(df_cm, annot=True)

from sklearn.metrics import classification_report

print(classification_report(labels, pred))

plt.figure(figsize = (25, 25))

### Modified for lisa
### changed index of images to show up
for i in range(30):
    plt.subplot(6, 5, i + 1)
    plt.grid(False)
    plt.xticks([])
    plt.yticks([])
    plt.rcParams["font.size"] = 16
    if i%10 == 0 :
        rndval = int(i/10)
    elif i%10 == 1 :
        rndval = 180 + int(i/10)
    elif i%10 == 2 :
        rndval = 470 + int(i/10)
    elif i%10 == 3 :
        rndval = 1470 + 0 + int(i/10)
    elif i%10 == 4 :
        rndval = 1470 + 3 + int(i/10)
    elif i%10 == 5 :
        rndval = 1470 + 6 + int(i/10)
    elif i%10 == 6 :
        rndval = 1470 + 9 + int(i/10)
    elif i%10 == 7 :
        rndval = 1470 + 12 + int(i/10)
    elif i%10 == 8 :
        rndval = 1470 + 15 + int(i/10)
    elif i%10 == 9 :
        rndval = 1470 + 18 + int(i/10)
    prediction = pred[rndval]
    actual = labels[rndval]
    col = 'g'
    if prediction != actual:
        col = 'r'
    plt.xlabel('Actual={} || Pred={}'.format(actual, prediction), color = col)
    for m in range(0, X_test[rndval].shape[0]): 
        for n in range(0, X_test[rndval].shape[1]): 
            tmpg = X_test[rndval][m][n][0]
            tmpr = X_test[rndval][m][n][2]
            X_test[rndval][m][n][0] = tmpr
            X_test[rndval][m][n][2] = tmpg
    plt.imshow(X_test[rndval])
plt.show()