Add Tensorflow Support to openvino_training_kit - Google Summer of Code

[leoheim]

Summary

Mentor: '@SHIVAMBASIA'

This PR adds comprehensive TensorFlow/Keras model wrappers with OpenVINO optimization for the GSoC 2025 "AI PC Model Training Kit" project. These wrappers enable developers to train, evaluate, and deploy deep learning models locally using OpenVINO's optimized backend and Intel Extension for TensorFlow (ITEX).

Key Features

• TensorFlow wrappers for classification, regression, segmentation, and detection
• Automatic OpenVINO optimization via Intel Extension for TensorFlow (ITEX)
• Quantization support: Post-Training Quantization (PTQ) and Quantization-Aware Training (QAT)
• Model persistence: TensorFlow SavedModel format with OpenVINO IR export
• Built-in metrics: accuracy, F1, precision, recall, IoU, Dice, mAP, etc.
• Inference modes: synchronous and asynchronous with OpenVINO Runtime

Supported Models

TensorFlow/Keras: Any tf.keras.Model (e.g., ResNet, UNet, MobileNet, custom architectures) via task-specific wrappers:

• ClassificationWrapper: Image/text classification with comprehensive metrics
• RegressionWrapper: Regression tasks with MSE, MAE, R² evaluation
• SegmentationWrapper: Semantic segmentation with IoU and Dice metrics
• DetectionWrapper: Object detection with custom metric support
• BaseWrapper: Core functionality for any TensorFlow model

How to Test

Basic Classification Example

import tensorflow as tf
from ov_training_kit.tensorflow import ClassificationWrapper

# Create model
model = tf.keras.Sequential([
    tf.keras.layers.Conv2D(32, 3, activation='relu', input_shape=(224, 224, 3)),
    tf.keras.layers.GlobalAveragePooling2D(),
    tf.keras.layers.Dense(10, activation='softmax')
])
model.compile(optimizer='adam', loss='sparse_categorical_crossentropy')

# Wrap and train
wrapper = ClassificationWrapper(model)
wrapper.fit(x_train, y_train, epochs=5, validation_data=(x_val, y_val))

# Evaluate with built-in metrics
accuracy = wrapper.evaluate_accuracy(x_test, y_test)
f1_score = wrapper.evaluate_f1(x_test, y_test, average='macro')
print(f"Accuracy: {accuracy:.3f}, F1: {f1_score:.3f}")