![]() #import librariesįrom sklearn.model_selection import train_test_splitįrom sklearn.linear_model import LogisticRegressionįrom sklearn.ensemble import RandomForestClassifierįrom trics import r2_score, mean_squared_error, confusion_matrix, precision_score, recall_score, accuracy_score, f1_scoreįrom sklearn.preprocessing import LabelEncoderįrom sklearn.preprocessing import MinMaxScaler Using cloud services like Google Cloud Auto ML definitely saves tons of time and avoid having to navigate the different image processing and extraction options. There is no one size fits all and the feature engineering should consider various feature extraction techniques and converting to feature vectors and will depend on your target use cases. The models can be refined and improved by providing more samples (full dataset is around 225MB), more features and combining both global and local features for increasing your model performance. I downloaded some images from the web and tried to predict and the model got most of it right with global features trained model, but pretty poor with the local features. ![]() Tried three ML algorithms: LogisticRegressor (LR), RandomForestClassifier (RFC) and Support Vector Machine(SVM), RFC performed best (close to 50% for raw pixels and 60% accuracy / precision for global features) but for local points of interest with ORB and BOVW, SVM had better performance. The focus was to extract the features and train the model and see how it performs with minimal tuning.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |