Image Classification (SVC & XGBoost)¶

Dataset

Imports¶

In [3]:

import os
import numpy as np

from skimage.io import imread
from skimage.color import rgb2gray
from skimage.transform import resize

from sklearn.model_selection import train_test_split, GridSearchCV
from sklearn.svm import SVC

from sklearn.metrics import classification_report
import pickle

from xgboost import XGBClassifier

import os import numpy as np from skimage.io import imread from skimage.color import rgb2gray from skimage.transform import resize from sklearn.model_selection import train_test_split, GridSearchCV from sklearn.svm import SVC from sklearn.metrics import classification_report import pickle from xgboost import XGBClassifier

Parking Lot¶

Load Data & Prepare¶

In [23]:

import kagglehub

# Download latest version
path = kagglehub.dataset_download("iasadpanwhar/parking-lot-detection-counter")

print("Path to dataset files:", path)

import kagglehub # Download latest version path = kagglehub.dataset_download("iasadpanwhar/parking-lot-detection-counter") print("Path to dataset files:", path)

Downloading from https://www.kaggle.com/api/v1/datasets/download/iasadpanwhar/parking-lot-detection-counter?dataset_version_number=1...

100%|██████████| 250M/250M [00:11<00:00, 21.9MB/s]

Extracting files...

Path to dataset files: /root/.cache/kagglehub/datasets/iasadpanwhar/parking-lot-detection-counter/versions/1

In [24]:

!mv /root/.cache/kagglehub/datasets/iasadpanwhar/parking-lot-detection-counter/versions/1/parking /content/

!mv /root/.cache/kagglehub/datasets/iasadpanwhar/parking-lot-detection-counter/versions/1/parking /content/

In [25]:

input_root_dir = "/content/parking"
input_dir = os.path.join(input_root_dir, "clf-data")
categories = ["empty", "not_empty"]

input_root_dir = "/content/parking" input_dir = os.path.join(input_root_dir, "clf-data") categories = ["empty", "not_empty"]

In [26]:

data = []
labels = []

for cat_idx, cat in enumerate(categories):
    for file_name in os.listdir(os.path.join(input_dir, cat)):
        img_path = os.path.join(input_dir, cat, file_name)
        img = imread(img_path)
        # print(img.shape) # (29, 60, 3)
        img = resize(img, (15, 15)) # 15, 30?
        data.append(img.flatten())
        labels.append(cat_idx)

data = [] labels = [] for cat_idx, cat in enumerate(categories): for file_name in os.listdir(os.path.join(input_dir, cat)): img_path = os.path.join(input_dir, cat, file_name) img = imread(img_path) # print(img.shape) # (29, 60, 3) img = resize(img, (15, 15)) # 15, 30? data.append(img.flatten()) labels.append(cat_idx)

Convert to numpy array

In [27]:

data = np.array(data)
labels = np.array(labels)

data = np.array(data) labels = np.array(labels)

Split train & test¶

In train_test_split, the stratify parameter ensures that the training and testing sets have the same proportion of classes (or labels) as the original dataset. This is particularly useful when dealing with imbalanced datasets, where some classes are underrepresented.

Read More

In [33]:

X_train_p, X_test_p, y_train_p, y_test_p = train_test_split(data, labels, test_size=0.2, random_state=42, stratify=labels)

X_train_p, X_test_p, y_train_p, y_test_p = train_test_split(data, labels, test_size=0.2, random_state=42, stratify=labels)

Train Model¶

SVC¶

In [34]:

classifier = SVC()

parameters = {"gamma": [0.1, 0.01, 0.001, 0.0001],
              "C": [1, 10, 100, 1000]}

grid_search = GridSearchCV(classifier, parameters)

classifier = SVC() parameters = {"gamma": [0.1, 0.01, 0.001, 0.0001], "C": [1, 10, 100, 1000]} grid_search = GridSearchCV(classifier, parameters)

In [35]:

%%time
grid_search.fit(X_train_p, y_train_p)

%%time grid_search.fit(X_train_p, y_train_p)

CPU times: user 1min 39s, sys: 693 ms, total: 1min 40s
Wall time: 1min 46s

Out[35]:

GridSearchCV(estimator=SVC(),
             param_grid={'C': [1, 10, 100, 1000],
                         'gamma': [0.1, 0.01, 0.001, 0.0001]})

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

Evaluate

In [36]:

best_estimator = grid_search.best_estimator_
best_estimator.get_params()

best_estimator = grid_search.best_estimator_ best_estimator.get_params()

Out[36]:

{'C': 1,
 'break_ties': False,
 'cache_size': 200,
 'class_weight': None,
 'coef0': 0.0,
 'decision_function_shape': 'ovr',
 'degree': 3,
 'gamma': 0.1,
 'kernel': 'rbf',
 'max_iter': -1,
 'probability': False,
 'random_state': None,
 'shrinking': True,
 'tol': 0.001,
 'verbose': False}

In [37]:

y_pred = best_estimator.predict(X_test_p)
print(classification_report(y_true=y_test_p, y_pred=y_pred))

y_pred = best_estimator.predict(X_test_p) print(classification_report(y_true=y_test_p, y_pred=y_pred))

              precision    recall  f1-score   support

           0       1.00      1.00      1.00       609
           1       1.00      1.00      1.00       609

    accuracy                           1.00      1218
   macro avg       1.00      1.00      1.00      1218
weighted avg       1.00      1.00      1.00      1218

XGB¶

In [2]:

%%time
xgb_classifier = XGBClassifier(device="gpu", random_state=42)

xgb_classifier.fit(X_train_p, y_train_p)

%%time xgb_classifier = XGBClassifier(device="gpu", random_state=42) xgb_classifier.fit(X_train_p, y_train_p)

In [39]:

y_pred = xgb_classifier.predict(X_test_p)
print(classification_report(y_true=y_test_p, y_pred=y_pred))

y_pred = xgb_classifier.predict(X_test_p) print(classification_report(y_true=y_test_p, y_pred=y_pred))

              precision    recall  f1-score   support

           0       1.00      1.00      1.00       609
           1       1.00      1.00      1.00       609

    accuracy                           1.00      1218
   macro avg       1.00      1.00      1.00      1218
weighted avg       1.00      1.00      1.00      1218

Save Model¶

In [ ]:

from google.colab import drive
drive.mount('/content/drive')

from google.colab import drive drive.mount('/content/drive')

Mounted at /content/drive

In [ ]:

pickle.dump(best_estimator, open("/content/drive/MyDrive/ML Models/parking_lot_detection_svc.pkl", "wb"))

pickle.dump(best_estimator, open("/content/drive/MyDrive/ML Models/parking_lot_detection_svc.pkl", "wb"))

Cat or Dog¶

Load Data¶

In [9]:

import kagglehub

# Download latest version
path = kagglehub.dataset_download("tongpython/cat-and-dog", force_download=True)

print("Path to dataset files:", path)

import kagglehub # Download latest version path = kagglehub.dataset_download("tongpython/cat-and-dog", force_download=True) print("Path to dataset files:", path)

Path to dataset files: /kaggle/input/cat-and-dog

In [10]:

!ls /kaggle/input/cat-and-dog

!ls /kaggle/input/cat-and-dog

test_set  training_set

In [11]:

input_root_dir = "/kaggle/input/cat-and-dog"
train_dir = os.path.join(input_root_dir, "training_set", "training_set")
test_dir = os.path.join(input_root_dir, "test_set", "test_set")
categories = ["cats", "dogs"]

input_root_dir = "/kaggle/input/cat-and-dog" train_dir = os.path.join(input_root_dir, "training_set", "training_set") test_dir = os.path.join(input_root_dir, "test_set", "test_set") categories = ["cats", "dogs"]

In [12]:

!ls {train_dir}

!ls {train_dir}

cats  dogs

Prepare Data¶

In [13]:

len(os.listdir(os.path.join(train_dir, "cats"))), len(os.listdir(os.path.join(test_dir, "cats")))

len(os.listdir(os.path.join(train_dir, "cats"))), len(os.listdir(os.path.join(test_dir, "cats")))

Out[13]:

(4001, 1012)

In [14]:

def process_img(img):
    if img.ndim == 3:
        img = rgb2gray(img)
    img = resize(img, (30, 30), anti_aliasing=True)
    return img.flatten()

def process_img(img): if img.ndim == 3: img = rgb2gray(img) img = resize(img, (30, 30), anti_aliasing=True) return img.flatten()

In [15]:

%%time
X_train = []
y_train = []

for cat_idx, cat in enumerate(categories):
    cat_dir = os.path.join(train_dir, cat)
    for file_name in os.listdir(cat_dir):
        if file_name.startswith('_'):  # Skip _DS_Store
            continue
        img_path = os.path.join(train_dir, cat, file_name)
        img = imread(img_path)
        img = process_img(img)
        X_train.append(img)
        y_train.append(cat_idx)

%%time X_train = [] y_train = [] for cat_idx, cat in enumerate(categories): cat_dir = os.path.join(train_dir, cat) for file_name in os.listdir(cat_dir): if file_name.startswith('_'): # Skip _DS_Store continue img_path = os.path.join(train_dir, cat, file_name) img = imread(img_path) img = process_img(img) X_train.append(img) y_train.append(cat_idx)

CPU times: user 1min 36s, sys: 1.88 s, total: 1min 38s
Wall time: 2min 8s

In [16]:

%%time
X_test = []
y_test = []

for cat_idx, cat in enumerate(categories):
    cat_dir = os.path.join(test_dir, cat)
    for file_name in os.listdir(cat_dir):
        if file_name.startswith('_'):  # Skip _DS_Store
            continue
        img_path = os.path.join(cat_dir, file_name)
        img = imread(img_path)
        img = process_img(img)  # You should normalize, resize, etc. here
        X_test.append(img)
        y_test.append(cat_idx)

%%time X_test = [] y_test = [] for cat_idx, cat in enumerate(categories): cat_dir = os.path.join(test_dir, cat) for file_name in os.listdir(cat_dir): if file_name.startswith('_'): # Skip _DS_Store continue img_path = os.path.join(cat_dir, file_name) img = imread(img_path) img = process_img(img) # You should normalize, resize, etc. here X_test.append(img) y_test.append(cat_idx)

CPU times: user 24.2 s, sys: 460 ms, total: 24.7 s
Wall time: 30.8 s

In [17]:

X_train = np.array(X_train)
y_train = np.array(y_train)

X_test = np.array(X_test)
y_test = np.array(y_test)

X_train = np.array(X_train) y_train = np.array(y_train) X_test = np.array(X_test) y_test = np.array(y_test)

Check balance

Train Model¶

In [29]:

def try_model(model):
  model.fit(X_train, y_train)
  y_pred = model.predict(X_test)
  print("Train\n", classification_report(y_true=y_train, y_pred=model.predict(X_train)))
  print("Test\n", classification_report(y_true=y_test, y_pred=y_pred), "\n")
  # print(model.get_params())

def try_model(model): model.fit(X_train, y_train) y_pred = model.predict(X_test) print("Train\n", classification_report(y_true=y_train, y_pred=model.predict(X_train))) print("Test\n", classification_report(y_true=y_test, y_pred=y_pred), "\n") # print(model.get_params())

SVC¶

In [90]:

%%time
svc_classifier = SVC(random_state=42)
try_model(svc_classifier)

%%time svc_classifier = SVC(random_state=42) try_model(svc_classifier)

Train
               precision    recall  f1-score   support

           0       0.82      0.77      0.79      4000
           1       0.78      0.83      0.81      4005

    accuracy                           0.80      8005
   macro avg       0.80      0.80      0.80      8005
weighted avg       0.80      0.80      0.80      8005

Test
               precision    recall  f1-score   support

           0       0.65      0.64      0.65      1011
           1       0.65      0.66      0.65      1012

    accuracy                           0.65      2023
   macro avg       0.65      0.65      0.65      2023
weighted avg       0.65      0.65      0.65      2023
 

CPU times: user 1min 35s, sys: 283 ms, total: 1min 35s
Wall time: 1min 36s

XGB¶

In [77]:

xgb_classifier = XGBClassifier(device="gpu", random_state=42)
try_model(xgb_classifier)

"""
Train
               precision    recall  f1-score   support

           0       1.00      1.00      1.00      4000
           1       1.00      1.00      1.00      4005

    accuracy                           1.00      8005
   macro avg       1.00      1.00      1.00      8005
weighted avg       1.00      1.00      1.00      8005

Test
               precision    recall  f1-score   support

           0       0.66      0.67      0.66      1011
           1       0.66      0.65      0.66      1012

    accuracy                           0.66      2023
   macro avg       0.66      0.66      0.66      2023
weighted avg       0.66      0.66      0.66      2023
"""

xgb_classifier = XGBClassifier(device="gpu", random_state=42) try_model(xgb_classifier) """ Train precision recall f1-score support 0 1.00 1.00 1.00 4000 1 1.00 1.00 1.00 4005 accuracy 1.00 8005 macro avg 1.00 1.00 1.00 8005 weighted avg 1.00 1.00 1.00 8005 Test precision recall f1-score support 0 0.66 0.67 0.66 1011 1 0.66 0.65 0.66 1012 accuracy 0.66 2023 macro avg 0.66 0.66 0.66 2023 weighted avg 0.66 0.66 0.66 2023 """

Train
               precision    recall  f1-score   support

           0       1.00      1.00      1.00      4000
           1       1.00      1.00      1.00      4005

    accuracy                           1.00      8005
   macro avg       1.00      1.00      1.00      8005
weighted avg       1.00      1.00      1.00      8005

Test
               precision    recall  f1-score   support

           0       0.66      0.67      0.66      1011
           1       0.66      0.65      0.66      1012

    accuracy                           0.66      2023
   macro avg       0.66      0.66      0.66      2023
weighted avg       0.66      0.66      0.66      2023
 

Out[77]:

'\nTrain\n               precision    recall  f1-score   support\n\n           0       1.00      1.00      1.00      4000\n           1       1.00      1.00      1.00      4005\n\n    accuracy                           1.00      8005\n   macro avg       1.00      1.00      1.00      8005\nweighted avg       1.00      1.00      1.00      8005\n\nTest\n               precision    recall  f1-score   support\n\n           0       0.66      0.67      0.66      1011\n           1       0.66      0.65      0.66      1012\n\n    accuracy                           0.66      2023\n   macro avg       0.66      0.66      0.66      2023\nweighted avg       0.66      0.66      0.66      2023\n'

In [78]:

# xgb_classifier = XGBClassifier(device="gpu", random_state=42, gamma=0.0001, eta=0.1, max_depth=4)
# try_model(xgb_classifier)

"""
Train
               precision    recall  f1-score   support

           0       0.87      0.79      0.83      4000
           1       0.81      0.88      0.84      4005

    accuracy                           0.84      8005
   macro avg       0.84      0.84      0.84      8005
weighted avg       0.84      0.84      0.84      8005

Test
               precision    recall  f1-score   support

           0       0.64      0.61      0.63      1011
           1       0.63      0.66      0.64      1012

    accuracy                           0.63      2023
   macro avg       0.63      0.63      0.63      2023
weighted avg       0.63      0.63      0.63      2023
"""

# xgb_classifier = XGBClassifier(device="gpu", random_state=42, gamma=0.0001, eta=0.1, max_depth=4) # try_model(xgb_classifier) """ Train precision recall f1-score support 0 0.87 0.79 0.83 4000 1 0.81 0.88 0.84 4005 accuracy 0.84 8005 macro avg 0.84 0.84 0.84 8005 weighted avg 0.84 0.84 0.84 8005 Test precision recall f1-score support 0 0.64 0.61 0.63 1011 1 0.63 0.66 0.64 1012 accuracy 0.63 2023 macro avg 0.63 0.63 0.63 2023 weighted avg 0.63 0.63 0.63 2023 """

Out[78]:

'\nTrain\n               precision    recall  f1-score   support\n\n           0       0.87      0.79      0.83      4000\n           1       0.81      0.88      0.84      4005\n\n    accuracy                           0.84      8005\n   macro avg       0.84      0.84      0.84      8005\nweighted avg       0.84      0.84      0.84      8005\n\nTest\n               precision    recall  f1-score   support\n\n           0       0.64      0.61      0.63      1011\n           1       0.63      0.66      0.64      1012\n\n    accuracy                           0.63      2023\n   macro avg       0.63      0.63      0.63      2023\nweighted avg       0.63      0.63      0.63      2023\n'

Save Model¶

In [76]:

from google.colab import drive
drive.mount('/content/drive')

from google.colab import drive drive.mount('/content/drive')

Mounted at /content/drive

In [91]:

pickle.dump(xgb_classifier, open("/content/drive/MyDrive/ML Models/cat_or_dog_xgb.pkl", "wb"))
pickle.dump(svc_classifier, open("/content/drive/MyDrive/ML Models/cat_or_dog_svc.pkl", "wb"))

pickle.dump(xgb_classifier, open("/content/drive/MyDrive/ML Models/cat_or_dog_xgb.pkl", "wb")) pickle.dump(svc_classifier, open("/content/drive/MyDrive/ML Models/cat_or_dog_svc.pkl", "wb"))

Prediction¶

In [47]:

from google.colab import files

from google.colab import files

In [85]:

uploaded = files.upload()

uploaded = files.upload()

Upload widget is only available when the cell has been executed in the current browser session. Please rerun this cell to enable.

Saving cat.jpg to cat (1).jpg

In [92]:

# Assuming you uploaded a single image
image_filename = next(iter(uploaded)) # Get the first (and likely only) filename
im = imread(image_filename)

# Assuming you uploaded a single image image_filename = next(iter(uploaded)) # Get the first (and likely only) filename im = imread(image_filename)

In [93]:

import matplotlib.pyplot as plt
plt.imshow(im)
plt.axis('off') # Hide axes for a cleaner display
plt.show()

import matplotlib.pyplot as plt plt.imshow(im) plt.axis('off') # Hide axes for a cleaner display plt.show()

In [94]:

# XGB
image = process_img(im)
y_pred = xgb_classifier.predict([image])
print(categories[y_pred[0]])

# XGB image = process_img(im) y_pred = xgb_classifier.predict([image]) print(categories[y_pred[0]])

cats

In [95]:

# SVC
image = process_img(im)
y_pred = svc_classifier.predict([image])
print(categories[y_pred[0]])

# SVC image = process_img(im) y_pred = svc_classifier.predict([image]) print(categories[y_pred[0]])

dogs