Classification-SIFT-BoW-SVM

Build a classifier to classsification transport using sift, bag of words and svm

Tree Project

Root/

data  # Data contains total 1854 image of 5 classes: bus, car, moto, pedestrian 

data_loader.py  # Imread image and label for data

sift_extractors.py # Extract features with SIFT and Build Bag of Word

build_model.py  # Build model SVM and training

gridSearchCV    # implement GridSearchCV find Hyper-params

Hyperparameters

• Number of centroid in Kmean: 60
• Penalty parameter C of the error term in SVM: 30

Label

• 'moto': 0
• 'car': 1
• 'pedestrian': 2
• 'truck': 3
• 'bus': 4

Statistic Data

- Statistics All Classes

- Statistics Classes

Accuracy

- Accuracy score: 72.9% - test size: 0.2 

Using GridSearchCV

- Best params

    'clf__C': 5,
    'clf__gamma': 0.01,
    'clf__kernel': 'rbf'
    
 - Accuracy:  75.72% +/-1.52%

Requirements

- opencv-contrib-python==3.4.2.16

- opencv-python==3.4.2.16