Image Data Augmentation Exercise : CIFAR-100

데이터가 더 부족한 세상으로!

여기 참고 - https://www.cs.toronto.edu/~kriz/cifar.html

Keras Update

!pip install keras-nightly

Collecting keras-nightly
  Downloading keras_nightly-3.7.0.dev2025010203-py3-none-any.whl.metadata (5.8 kB)
Requirement already satisfied: absl-py in /usr/local/lib/python3.10/dist-packages (from keras-nightly) (1.4.0)
Requirement already satisfied: numpy in /usr/local/lib/python3.10/dist-packages (from keras-nightly) (1.26.4)
Requirement already satisfied: rich in /usr/local/lib/python3.10/dist-packages (from keras-nightly) (13.9.4)
Requirement already satisfied: namex in /usr/local/lib/python3.10/dist-packages (from keras-nightly) (0.0.8)
Requirement already satisfied: h5py in /usr/local/lib/python3.10/dist-packages (from keras-nightly) (3.12.1)
Requirement already satisfied: optree in /usr/local/lib/python3.10/dist-packages (from keras-nightly) (0.13.1)
Requirement already satisfied: ml-dtypes in /usr/local/lib/python3.10/dist-packages (from keras-nightly) (0.4.1)
Requirement already satisfied: packaging in /usr/local/lib/python3.10/dist-packages (from keras-nightly) (24.2)
Requirement already satisfied: typing-extensions>=4.5.0 in /usr/local/lib/python3.10/dist-packages (from optree->keras-nightly) (4.12.2)
Requirement already satisfied: markdown-it-py>=2.2.0 in /usr/local/lib/python3.10/dist-packages (from rich->keras-nightly) (3.0.0)
Requirement already satisfied: pygments<3.0.0,>=2.13.0 in /usr/local/lib/python3.10/dist-packages (from rich->keras-nightly) (2.18.0)
Requirement already satisfied: mdurl~=0.1 in /usr/local/lib/python3.10/dist-packages (from markdown-it-py>=2.2.0->rich->keras-nightly) (0.1.2)
Downloading keras_nightly-3.7.0.dev2025010203-py3-none-any.whl (1.3 MB)
   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 1.3/1.3 MB 17.4 MB/s eta 0:00:00
Installing collected packages: keras-nightly
Successfully installed keras-nightly-3.7.0.dev2025010203

Data Loading

import numpy as np
import matplotlib.pyplot as plt

from keras.datasets.cifar100 import load_data

(train_x, train_y), (test_x, test_y) = load_data()
# (train_x, train_y), (test_x, test_y) = load_data(label_mode='coarse')

np.unique(train_y)

array([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15, 16,
       17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33,
       34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50,
       51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67,
       68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84,
       85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99])

label_dict = {0:'apple', 1: 'aquarium_fish', 2: 'baby', 3: 'bear', 4: 'beaver', 5: 'bed', 6: 'bee', 7: 'beetle', 8: 'bicycle', 9: 'bottle',
              10: 'bowl', 11: 'boy',12: 'bridge',13: 'bus',14: 'butterfly',15: 'camel',16: 'can',17: 'castle',18: 'caterpillar',19: 'cattle',
              20: 'chair',21: 'chimpanzee',22: 'clock',23: 'cloud',24: 'cockroach',25: 'couch',26: 'cra',27: 'crocodile',28: 'cup',29: 'dinosaur',
              30: 'dolphin',31: 'elephant',32: 'flatfish',33: 'forest',34: 'fox',35: 'girl',36: 'hamster',37: 'house',38: 'kangaroo',39: 'keyboard',
              40: 'lamp',41: 'lawn_mower',42: 'leopard',43: 'lion',44: 'lizard',45: 'lobster',46: 'man',47: 'maple_tree',48: 'motorcycle',49: 'mountain',
              50: 'mouse',51: 'mushroom',52: 'oak_tree',53: 'orange',54: 'orchid',55: 'otter',56: 'palm_tree',57: 'pear',58: 'pickup_truck',59: 'pine_tree',
              60: 'plain',61: 'plate',62: 'poppy',63: 'porcupine',64: 'possum',65: 'rabbit',66: 'raccoon',67: 'ray',68: 'road',69: 'rocket',
              70: 'rose',71: 'sea',72: 'seal',73: 'shark',74: 'shrew',75: 'skunk',76: 'skyscraper',77: 'snail',78: 'snake',79: 'spider',
              80: 'squirrel',81: 'streetcar',82: 'sunflower',83: 'sweet_pepper',84: 'table',85: 'tank',86: 'telephone',87: 'television',88: 'tiger',89: 'tractor',
              90: 'train',91: 'trout',92: 'tulip',93: 'turtle',94: 'wardrobe',95: 'whale',96: 'willow_tree',97: 'wolf',98: 'woman',99: 'worm'
            }

label_dict[0]

apple

데이터 살펴보기

rand_i = np.random.randint(0, train_x.shape[0])

plt.title(f'idx: {rand_i} , class: { label_dict[train_y[rand_i][0]] }')
plt.imshow( train_x[rand_i] )
plt.show()

rows = 5
fig, axes = plt.subplots(rows, len(label_dict), figsize=(len(label_dict), rows) )

for img_id in range(len(label_dict)) :
    imgs = train_x[train_y.reshape(-1)==img_id]
    imgs_len = len(imgs)

    for row_i in range(rows) :
        axe = axes[row_i, img_id]
        axe.imshow( imgs[np.random.randint(imgs_len)], interpolation='none' )
        axe.axis('off')

plt.tight_layout()
plt.show()

Data Preprocessing

Data split
- training set : validation set = 8 : 2
- 재현을 위한 난수 고정 : 2024

from sklearn.model_selection import train_test_split as tts

train_x, val_x, train_y, val_y = tts(train_x, train_y, test_size=0.2, random_state=2024)

train_x.shape, train_y.shape

((40000, 32, 32, 3), (40000, 1))

Scaling
- min-max scaling (선택사항)
  1. RGB 정보 전체를 min-max
  2. R 따로 G 따로 B 따로 min-max, 그 후 하나로 통합

# min-max scaling 1번 방법
max_n, min_n = train_x.max(), train_x.min()
max_n, min_n

(255, 0)

train_x_mm1 = (train_x - min_n) / (max_n - min_n)
val_x_mm1 = (val_x - min_n) / (max_n - min_n)
test_x_mm1 = (test_x - min_n) / (max_n - min_n)

train_x_mm1.max(), train_x_mm1.min()

(1.0, 0.0)

# min-max scaling 2번 방법
tr_r_max, tr_r_min = train_x[:,:,:,0].max(), train_x[:,:,:,0].min()
tr_g_max, tr_g_min = train_x[:,:,:,1].max(), train_x[:,:,:,1].min()
tr_b_max, tr_b_min = train_x[:,:,:,2].max(), train_x[:,:,:,2].min()

train_r_mm = (train_x[:,:,:,0] - tr_r_min) / (tr_r_max - tr_r_min)
train_g_mm = (train_x[:,:,:,1] - tr_g_min) / (tr_g_max - tr_g_min)
train_b_mm = (train_x[:,:,:,2] - tr_b_min) / (tr_b_max - tr_b_min)

train_x_mm = np.stack((train_r_mm, train_g_mm, train_b_mm), axis=3)

train_x_mm.shape

(40000, 32, 32, 3)

val_r_mm = (val_x[:,:,:,0] - tr_r_min) / (tr_r_max - tr_r_min)
val_g_mm = (val_x[:,:,:,1] - tr_g_min) / (tr_g_max - tr_g_min)
val_b_mm = (val_x[:,:,:,2] - tr_b_min) / (tr_b_max - tr_b_min)

val_x_mm = np.stack((val_r_mm, val_g_mm, val_b_mm), axis=3)

val_x_mm.shape

(10000, 32, 32, 3)

test_r_mm = (test_x[:,:,:,0] - tr_r_min) / (tr_r_max - tr_r_min)
test_g_mm = (test_x[:,:,:,1] - tr_g_min) / (tr_g_max - tr_g_min)
test_b_mm = (test_x[:,:,:,2] - tr_b_min) / (tr_b_max - tr_b_min)

test_x_mm = np.stack((test_r_mm, test_g_mm, test_b_mm), axis=3)

test_x_mm.shape

(10000, 32, 32, 3)

One-hot encoding

train_y.shape
train_y

array([[ 2],
       [58],
       [74],
       ...,
       [15],
       [78],
       [35]])

class_n = len(np.unique(train_y))

from keras.utils import to_categorical

train_y = to_categorical(train_y, class_n)
val_y = to_categorical(val_y, class_n)
test_y = to_categorical(test_y, class_n)

Data shape 재확인

train_x_mm.shape, train_y.shape

((40000, 32, 32, 3), (40000, 100))

train_y[0]

array([0., 0., 1., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
       0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
       0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
       0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
       0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
       0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.])

Modeling : CNN

조건
1. Sequential API, Functiona API 중 택일.
2. Image Augmentation Layer를 최소 하나 이상 넣을 것! : 공식 문서 참고
3. 이 구조를 미니 버전으로 활용해봐도 좋다.
4. DropOut, BatchNormalization 등의 기능도 같이 활용해보자.
5. Early Stopping을 사용할 것.

import keras

from keras.backend import clear_session
from keras.models import Model
from keras.layers import Input, Dense, Flatten, BatchNormalization, Dropout, Conv2D, MaxPool2D
from keras.layers import RandomRotation, RandomTranslation, RandomZoom, RandomFlip

# Functional API
# 1. 세션 클리어
clear_session()

# 2. 레이어 엮기
il = Input(shape=(32,32,3) )

al = RandomRotation(0.2)(il)
al = RandomTranslation(0.2, 0.2)(al)
al = RandomZoom(0.1)(al)
al = RandomFlip()(al)

hl = Conv2D(filters=64,        # 새롭게 제작하려는 feature map의 수! 서로 다른 filter의 수
            kernel_size=(3,3), # Conv filter의 가로세로 사이즈
            strides=(1,1),     # Conv filter의 이동 보폭
            padding='same',    # 1. 이전 feature map 사이즈 유지 | 2. 외곽 정보 더 반영
            activation='relu'  # 명시 주의!
            )(al)
hl = Conv2D(filters=64,        # 새롭게 제작하려는 feature map의 수! 서로 다른 filter의 수
            kernel_size=(3,3), # Conv filter의 가로세로 사이즈
            strides=(1,1),     # Conv filter의 이동 보폭
            padding='same',    # 1. 이전 feature map 사이즈 유지 | 2. 외곽 정보 더 반영
            activation='relu'  # 명시 주의!
            )(hl)
hl = MaxPool2D(pool_size=(2,2),# Pool filter 가로세로 사이즈
               strides=(2,2)   # Pool filter의 이동 보폭
               )(hl)
hl = BatchNormalization()(hl)
hl = Dropout(0.5)(hl)

hl = Conv2D(filters=128,        # 새롭게 제작하려는 feature map의 수! 서로 다른 filter의 수
            kernel_size=(3,3), # Conv filter의 가로세로 사이즈
            strides=(1,1),     # Conv filter의 이동 보폭
            padding='same',    # 1. 이전 feature map 사이즈 유지 | 2. 외곽 정보 더 반영
            activation='relu'  # 명시 주의!
            )(hl)
hl = Conv2D(filters=128,        # 새롭게 제작하려는 feature map의 수! 서로 다른 filter의 수
            kernel_size=(3,3), # Conv filter의 가로세로 사이즈
            strides=(1,1),     # Conv filter의 이동 보폭
            padding='same',    # 1. 이전 feature map 사이즈 유지 | 2. 외곽 정보 더 반영
            activation='relu'  # 명시 주의!
            )(hl)
hl = MaxPool2D(pool_size=(2,2),# Pool filter 가로세로 사이즈
               strides=(2,2)   # Pool filter의 이동 보폭
               )(hl)
hl = BatchNormalization()(hl)
hl = Dropout(0.5)(hl)

hl = Flatten()(hl)
hl = Dense(1024, activation='relu')(hl)
ol = Dense(100, activation='softmax')(hl)

# 3. 모델 시작과 끝 지정
model = Model(il, ol)

# 4. 컴파일
model.compile(optimizer='adam', loss=keras.losses.categorical_crossentropy,
              metrics=['accuracy'])

model.summary()

Early Stopping

from keras.callbacks import EarlyStopping

es = EarlyStopping(monitor='val_loss',       # 얼리스토핑 적용할 관측 대상
                   min_delta=0,              # 임계값.
                   patience=3,               # 성능 개선 미발생시 몇 epoch 더 진행할 것인가.
                   verbose=1,
                   restore_best_weights=True # 가장 성능 좋은 epoch의 가중치로 되돌림
                   )

.fit( )

model.fit(train_x_mm, train_y,
          validation_data=(val_x_mm, val_y),
          epochs=10000, verbose=1,
          callbacks=[es]                # 얼리스토핑 적용!
          )

Epoch 1/10000
1250/1250 ━━━━━━━━━━━━━━━━━━━━ 31s 19ms/step - accuracy: 0.0456 - loss: 4.5369 - val_accuracy: 0.0940 - val_loss: 3.9263
Epoch 2/10000
1250/1250 ━━━━━━━━━━━━━━━━━━━━ 23s 18ms/step - accuracy: 0.0948 - loss: 3.9055 - val_accuracy: 0.1098 - val_loss: 3.7984
Epoch 3/10000
1250/1250 ━━━━━━━━━━━━━━━━━━━━ 42s 19ms/step - accuracy: 0.1158 - loss: 3.7538 - val_accuracy: 0.1370 - val_loss: 3.6270
Epoch 4/10000
1250/1250 ━━━━━━━━━━━━━━━━━━━━ 40s 18ms/step - accuracy: 0.1286 - loss: 3.6649 - val_accuracy: 0.1374 - val_loss: 3.7671
Epoch 5/10000
1250/1250 ━━━━━━━━━━━━━━━━━━━━ 41s 18ms/step - accuracy: 0.1441 - loss: 3.5871 - val_accuracy: 0.1475 - val_loss: 3.7322
Epoch 6/10000
1250/1250 ━━━━━━━━━━━━━━━━━━━━ 41s 18ms/step - accuracy: 0.1539 - loss: 3.5368 - val_accuracy: 0.1592 - val_loss: 3.5241
Epoch 7/10000
1250/1250 ━━━━━━━━━━━━━━━━━━━━ 23s 19ms/step - accuracy: 0.1653 - loss: 3.4803 - val_accuracy: 0.1831 - val_loss: 3.4410
Epoch 8/10000
1250/1250 ━━━━━━━━━━━━━━━━━━━━ 41s 18ms/step - accuracy: 0.1703 - loss: 3.4352 - val_accuracy: 0.1794 - val_loss: 3.4417
Epoch 9/10000
1250/1250 ━━━━━━━━━━━━━━━━━━━━ 41s 18ms/step - accuracy: 0.1837 - loss: 3.3873 - val_accuracy: 0.1937 - val_loss: 3.3617
Epoch 10/10000
1250/1250 ━━━━━━━━━━━━━━━━━━━━ 42s 20ms/step - accuracy: 0.1846 - loss: 3.3526 - val_accuracy: 0.1812 - val_loss: 3.4366
Epoch 11/10000
1250/1250 ━━━━━━━━━━━━━━━━━━━━ 40s 19ms/step - accuracy: 0.1890 - loss: 3.3301 - val_accuracy: 0.1740 - val_loss: 3.4759
Epoch 12/10000
1250/1250 ━━━━━━━━━━━━━━━━━━━━ 40s 18ms/step - accuracy: 0.1934 - loss: 3.2969 - val_accuracy: 0.2229 - val_loss: 3.1815
Epoch 13/10000
1250/1250 ━━━━━━━━━━━━━━━━━━━━ 41s 18ms/step - accuracy: 0.1997 - loss: 3.2570 - val_accuracy: 0.2010 - val_loss: 3.3584
Epoch 14/10000
1250/1250 ━━━━━━━━━━━━━━━━━━━━ 41s 18ms/step - accuracy: 0.2046 - loss: 3.2396 - val_accuracy: 0.2056 - val_loss: 3.2955
Epoch 15/10000
1250/1250 ━━━━━━━━━━━━━━━━━━━━ 42s 19ms/step - accuracy: 0.2085 - loss: 3.2195 - val_accuracy: 0.1944 - val_loss: 3.4265
Epoch 15: early stopping
Restoring model weights from the end of the best epoch: 12.
<keras.src.callbacks.history.History at 0x781b022d8b80>

.evaluate( )

model.evaluate(test_x_mm, test_y)

313/313 ━━━━━━━━━━━━━━━━━━━━ 1s 5ms/step - accuracy: 0.2255 - loss: 3.1848

[3.178687810897827, 0.22519999742507935]

.predict( )

y_pred = model.predict(test_x_mm)

313/313 ━━━━━━━━━━━━━━━━━━━━ 1s 3ms/step

# 원핫 인코딩 한 것을 다시 묶어주는 코드
# 평가 지표 및 실제 데이터 확인을 위해 필요

y_pred_arg = np.argmax(y_pred, axis=1)
test_y_arg = np.argmax(test_y, axis=1)

y_pred_arg[0]

test_y_arg[0]

평가 지표

from sklearn.metrics import accuracy_score, classification_report

accuracy_score(test_y_arg, y_pred_arg)

0.2252

print( classification_report(test_y_arg, y_pred_arg, target_names=list(label_dict.values())) )

               precision    recall  f1-score   support

        apple       0.59      0.46      0.52       100
aquarium_fish       0.33      0.30      0.31       100
         baby       0.47      0.22      0.30       100
         bear       0.09      0.03      0.05       100
       beaver       0.14      0.03      0.05       100
          bed       0.00      0.00      0.00       100
          bee       0.19      0.30      0.23       100
       beetle       0.16      0.31      0.21       100
      bicycle       0.20      0.10      0.13       100
       bottle       0.40      0.16      0.23       100
         bowl       0.00      0.00      0.00       100
          boy       0.17      0.04      0.06       100
       bridge       0.23      0.10      0.14       100
          bus       0.10      0.05      0.07       100
    butterfly       0.14      0.19      0.16       100
        camel       0.18      0.19      0.19       100
          can       0.19      0.17      0.18       100
       castle       0.28      0.25      0.26       100
  caterpillar       0.31      0.27      0.29       100
       cattle       0.11      0.04      0.06       100
        chair       0.14      0.55      0.22       100
   chimpanzee       0.33      0.40      0.36       100
        clock       0.11      0.23      0.15       100
        cloud       0.55      0.52      0.53       100
    cockroach       0.27      0.62      0.37       100
        couch       0.00      0.00      0.00       100
          cra       0.20      0.11      0.14       100
    crocodile       0.00      0.00      0.00       100
          cup       0.10      0.03      0.05       100
     dinosaur       0.14      0.09      0.11       100
      dolphin       0.34      0.25      0.29       100
     elephant       0.15      0.24      0.19       100
     flatfish       0.17      0.07      0.10       100
       forest       0.21      0.30      0.25       100
          fox       0.22      0.10      0.14       100
         girl       0.15      0.28      0.20       100
      hamster       0.23      0.35      0.28       100
        house       0.30      0.18      0.22       100
     kangaroo       0.15      0.13      0.14       100
     keyboard       0.18      0.38      0.25       100
         lamp       0.50      0.04      0.07       100
   lawn_mower       0.42      0.43      0.42       100
      leopard       0.14      0.17      0.15       100
         lion       0.12      0.52      0.20       100
       lizard       0.18      0.07      0.10       100
      lobster       0.15      0.16      0.16       100
          man       0.26      0.16      0.20       100
   maple_tree       0.29      0.32      0.31       100
   motorcycle       0.16      0.55      0.24       100
     mountain       0.58      0.14      0.23       100
        mouse       0.00      0.00      0.00       100
     mushroom       0.13      0.20      0.16       100
     oak_tree       0.34      0.79      0.47       100
       orange       0.36      0.72      0.48       100
       orchid       0.32      0.51      0.39       100
        otter       0.00      0.00      0.00       100
    palm_tree       0.47      0.19      0.27       100
         pear       0.33      0.15      0.21       100
 pickup_truck       0.12      0.19      0.15       100
    pine_tree       0.11      0.08      0.09       100
        plain       0.77      0.30      0.43       100
        plate       0.21      0.36      0.26       100
        poppy       0.26      0.65      0.37       100
    porcupine       0.13      0.09      0.11       100
       possum       0.08      0.02      0.03       100
       rabbit       0.07      0.03      0.04       100
      raccoon       0.23      0.07      0.11       100
          ray       0.41      0.32      0.36       100
         road       0.78      0.35      0.48       100
       rocket       0.42      0.36      0.39       100
         rose       0.35      0.34      0.35       100
          sea       0.44      0.22      0.29       100
         seal       0.00      0.00      0.00       100
        shark       0.33      0.38      0.35       100
        shrew       0.09      0.05      0.06       100
        skunk       0.24      0.59      0.34       100
   skyscraper       0.53      0.36      0.43       100
        snail       0.36      0.04      0.07       100
        snake       0.06      0.20      0.09       100
       spider       0.29      0.04      0.07       100
     squirrel       0.00      0.00      0.00       100
    streetcar       0.14      0.10      0.12       100
    sunflower       0.49      0.84      0.62       100
 sweet_pepper       0.33      0.16      0.21       100
        table       0.32      0.07      0.11       100
         tank       0.13      0.27      0.18       100
    telephone       0.22      0.45      0.30       100
   television       0.16      0.11      0.13       100
        tiger       0.11      0.54      0.18       100
      tractor       0.28      0.32      0.30       100
        train       0.16      0.10      0.12       100
        trout       0.33      0.01      0.02       100
        tulip       0.13      0.04      0.06       100
       turtle       0.21      0.18      0.19       100
     wardrobe       0.46      0.22      0.30       100
        whale       0.39      0.37      0.38       100
  willow_tree       0.31      0.15      0.20       100
         wolf       0.15      0.13      0.14       100
        woman       0.21      0.11      0.14       100
         worm       0.15      0.15      0.15       100

     accuracy                           0.23     10000
    macro avg       0.24      0.23      0.20     10000
 weighted avg       0.24      0.23      0.20     10000

/usr/local/lib/python3.10/dist-packages/sklearn/metrics/_classification.py:1565: UndefinedMetricWarning: Precision is ill-defined and being set to 0.0 in labels with no predicted samples. Use `zero_division` parameter to control this behavior.
  _warn_prf(average, modifier, f"{metric.capitalize()} is", len(result))
/usr/local/lib/python3.10/dist-packages/sklearn/metrics/_classification.py:1565: UndefinedMetricWarning: Precision is ill-defined and being set to 0.0 in labels with no predicted samples. Use `zero_division` parameter to control this behavior.
  _warn_prf(average, modifier, f"{metric.capitalize()} is", len(result))
/usr/local/lib/python3.10/dist-packages/sklearn/metrics/_classification.py:1565: UndefinedMetricWarning: Precision is ill-defined and being set to 0.0 in labels with no predicted samples. Use `zero_division` parameter to control this behavior.
  _warn_prf(average, modifier, f"{metric.capitalize()} is", len(result))

Visualization

실제 데이터 확인

rand_idx = np.random.randint(0, len(y_pred_arg))
test_idx = test_y_arg[rand_idx]
pred_idx = y_pred_arg[rand_idx]
class_prob = np.floor( y_pred[rand_idx]*100 )

print(f'idx = {rand_idx}')
print(f'해당 인덱스의 이미지는 {label_dict[test_idx]}')
print(f'모델의 예측 : {label_dict[pred_idx]}')
print(f'모델의 클래스별 확률 : ')
print('-------------------')
for idx, val in enumerate( list(label_dict.values()) ) :
    print(val, class_prob[idx])
print('=================================================')

if test_y_arg[rand_idx] == y_pred_arg[rand_idx] :
    print('정답')
else :
    print('땡')

plt.imshow(test_x[rand_idx])
plt.show()

idx = 1630
해당 인덱스의 이미지는 orchid
모델의 예측 : orchid
모델의 클래스별 확률 : 
-------------------
apple 0.0
aquarium_fish 0.0
baby 0.0
bear 0.0
beaver 0.0
bed 0.0
bee 0.0
beetle 0.0
bicycle 0.0
bottle 0.0
bowl 0.0
boy 0.0
bridge 0.0
bus 0.0
butterfly 0.0
camel 0.0
can 0.0
castle 0.0
caterpillar 0.0
cattle 0.0
chair 0.0
chimpanzee 0.0
clock 0.0
cloud 0.0
cockroach 0.0
couch 0.0
cra 0.0
crocodile 0.0
cup 0.0
dinosaur 0.0
dolphin 0.0
elephant 0.0
flatfish 0.0
forest 0.0
fox 0.0
girl 0.0
hamster 0.0
house 0.0
kangaroo 0.0
keyboard 0.0
lamp 0.0
lawn_mower 0.0
leopard 0.0
lion 0.0
lizard 0.0
lobster 0.0
man 0.0
maple_tree 0.0
motorcycle 0.0
mountain 0.0
mouse 0.0
mushroom 0.0
oak_tree 0.0
orange 0.0
orchid 66.0
otter 0.0
palm_tree 0.0
pear 0.0
pickup_truck 0.0
pine_tree 0.0
plain 0.0
plate 0.0
poppy 3.0
porcupine 0.0
possum 0.0
rabbit 0.0
raccoon 0.0
ray 0.0
road 0.0
rocket 0.0
rose 19.0
sea 0.0
seal 0.0
shark 0.0
shrew 0.0
skunk 0.0
skyscraper 0.0
snail 0.0
snake 0.0
spider 0.0
squirrel 0.0
streetcar 0.0
sunflower 0.0
sweet_pepper 0.0
table 0.0
tank 0.0
telephone 0.0
television 0.0
tiger 0.0
tractor 0.0
train 0.0
trout 0.0
tulip 8.0
turtle 0.0
wardrobe 0.0
whale 0.0
willow_tree 0.0
wolf 0.0
woman 0.0
worm 0.0
=================================================
정답

틀린 이미지만 확인해보기

temp = (test_y_arg == y_pred_arg)
false_idx = np.where(temp==False)[0]
false_len = len(false_idx)
false_len

rand_idx = false_idx[np.random.randint(0, false_len)]
test_idx = test_y_arg[rand_idx]
pred_idx = y_pred_arg[rand_idx]
class_prob = np.floor( y_pred[rand_idx]*100 )

print(f'idx = {rand_idx}')
print(f'해당 인덱스의 이미지는 {label_dict[test_idx]}')
print(f'모델의 예측 : {label_dict[pred_idx]}')
print(f'모델의 클래스별 확률 : ')
print('-------------------')
for idx, val in enumerate( list(label_dict.values()) ) :
    print(val, class_prob[idx])
print('=================================================')

if test_y_arg[rand_idx] == y_pred_arg[rand_idx] :
    print('정답')
else :
    print('땡')

plt.imshow(test_x[rand_idx] )
plt.show()

idx = 1471
해당 인덱스의 이미지는 couch
모델의 예측 : streetcar
모델의 클래스별 확률 : 
-------------------
apple 0.0
aquarium_fish 0.0
baby 1.0
bear 1.0
beaver 1.0
bed 1.0
bee 1.0
beetle 2.0
bicycle 0.0
bottle 1.0
bowl 1.0
boy 1.0
bridge 0.0
bus 3.0
butterfly 3.0
camel 0.0
can 3.0
castle 0.0
caterpillar 0.0
cattle 2.0
chair 0.0
chimpanzee 2.0
clock 1.0
cloud 0.0
cockroach 2.0
couch 1.0
cra 1.0
crocodile 0.0
cup 0.0
dinosaur 0.0
dolphin 0.0
elephant 1.0
flatfish 2.0
forest 0.0
fox 0.0
girl 1.0
hamster 0.0
house 1.0
kangaroo 0.0
keyboard 0.0
lamp 0.0
lawn_mower 1.0
leopard 0.0
lion 0.0
lizard 0.0
lobster 2.0
man 0.0
maple_tree 0.0
motorcycle 3.0
mountain 0.0
mouse 0.0
mushroom 0.0
oak_tree 0.0
orange 0.0
orchid 0.0
otter 1.0
palm_tree 0.0
pear 0.0
pickup_truck 6.0
pine_tree 0.0
plain 0.0
plate 0.0
poppy 0.0
porcupine 0.0
possum 0.0
rabbit 0.0
raccoon 0.0
ray 0.0
road 0.0
rocket 0.0
rose 0.0
sea 0.0
seal 0.0
shark 0.0
shrew 0.0
skunk 0.0
skyscraper 0.0
snail 0.0
snake 1.0
spider 0.0
squirrel 0.0
streetcar 6.0
sunflower 0.0
sweet_pepper 0.0
table 1.0
tank 0.0
telephone 1.0
television 3.0
tiger 1.0
tractor 3.0
train 3.0
trout 0.0
tulip 0.0
turtle 0.0
wardrobe 0.0
whale 0.0
willow_tree 0.0
wolf 0.0
woman 1.0
worm 0.0
=================================================
땡

'AI_딥 러닝_시각지능' 카테고리의 다른 글

AI_파이썬_시각지능_CNN_남이 만든 CNN모델(Pretrained ) (0)	2025.01.02
AI_파이썬_시각지능_CNN_MyModel_Save&Load_2024 (0)	2025.01.02
AI_파이썬_시각지능_CNN_SmallSize_notMNIST_2024 (0)	2024.12.30
AI_파이썬_시각지능_CNN_Image_SmallData2024 (0)	2024.12.30
AI_파이썬_시각지능_CNN_exercise_notMNIST_small_2024 (0)	2024.12.30

T 스토리 블로그

AI_파이썬_시각지능_CNN_SmallSize_ImageData_2024

Image Data Augmentation Exercise : CIFAR-100

데이터가 더 부족한 세상으로!

Keras Update

Data Loading

Data Preprocessing

Modeling : CNN

Visualization

'AI_딥 러닝_시각지능' 카테고리의 다른 글

티스토리툴바

AI_파이썬_시각지능_CNN_SmallSize_ImageData_2024

Image Data Augmentation Exercise : CIFAR-100

데이터가 더 부족한 세상으로!

Keras Update

Data Loading

Data Preprocessing

Modeling : CNN

Visualization

'AI_딥 러닝_시각지능' 카테고리의 다른 글

'AI_딥 러닝_시각지능' Related Articles

티스토리툴바