DeepLearning-ImageClassification-Toolkit
End-to-end Image Classification using Deep Learning toolkit for custom image datasets. Features include Pre-Processing, Training with Multiple CNN Architectures and Statistical Inference Tools. Special utilities for RAM optimization, Learning Rate Scheduling, Detailed Code Comments and Necessary Diagrams are included. This Codebase requires TensorFlow, kindly make sure you have that installed.
Overview of Complete System :
PreProcessing Major Highlights:
- Remove Background and Extract Object from Source Image
- Convert Image into NumPy Array
- Split Data into Training, Testing and Validation Data
- Create One-Hot-Encoding for Categorical Labels
- Provided Functionalities for Oversampling
- Provided Comments for Better Understanding
- More Features and Detailed Explanation Available in Code
Models Trained:
- EfficientNetB0
- InceptionResNetV2
- ResNet50
- VGG16
- Added Tensorflow strategy.scope() to Distribute the Training of Model on All Available GPUs
Statistical Analysis for:
- Generate Graphs for Validation Loss, Validation Accuracy, F-1 Score, Validation AUC
- Generate Confusion Matrix for Each Model