removed duplicate function | Improved CNN-Model; Added 'prediction.py'

[ModusMorris]

@ModusMorris:

Removed duplicate is_video_in_csv() from generator.py.
Now imports is_video_in_csv() from metadata_handler.py instead.
Maintains clean, modular, and reusable code

---

@AlhaririAnas:

Experiments with Various Models

I experimented with various models, and the best results were achieved with the ResNet1DInspired model, currently named StepCounterCNN.

This model is based on a 1D Convolutional Neural Network (CNN) architecture with residual blocks, which help in efficiently training. The architecture consists of:

1. An initial convolutional layer with batch normalization and max pooling.
2. Two residual blocks, each comprising two convolutional layers with batch normalization, ReLU activation, and dropout.
3. Fully connected layers for classification, followed by a sigmoid activation for binary output.

New results:

New File: prediction.py

I created the prediction.py file. When this file is executed, it accesses the trained model and makes predictions. I integrated this functionality into training.py to make the code more organized.

Improving Generalization

To improve the model's generalization, I added slight noise to the training data. This has shown to enhance learning and reduce the risk of overfitting.

If you have any suggestions for improvement or feedback, please let me know.

[ModusMorris]

I think the results are great, so we definitely have very meaningful content to work with