Author: Sarang Galada
Email ID: sarang.galada@gmail.com
Date created: 26/12/2023
Description: Classification of Bow (weapon) images scraped from the web into 5 types using Transfer-learning and Fine-tuning of 3 popular CNNs - ResNet50V2, InceptionV3 and DenseNet121. Finally the best performing fine-tuned models of each are ensembled with majority-rule voting
Results: Achieved a maximum accuracy and weighted-F1-score of 0.91, with 10 epochs of training on a bootstrapped dataset of 810 images
- Problem:
Image Classification - Data:
Bow (weapon) Images - Models:
ResNet50V2,InceptionV3,DenseNet121 - Key libraries used:
TensorFlow Keras,PIL,Scikit-Learn
Note: For testing and demo, see or run Code/Ensemble123.ipynb