Vision Forge

Video Demo: https://www.youtube.com/watch?v=QUqdwoElxEI

Description:

Vision Forge is an intuitive application designed for creating, training, and utilizing image classification models through a user-friendly graphical user interface (GUI). This app eliminates the need for any coding, making it accessible to users of all skill levels.

Features:

• Custom Model Creation: Customize your image classification models with options to select pretrained models, optimizers, epochs, hidden units/neurons, and learning rate.
• Data Loading: Easily load your dataset, and the app will automatically infer the number of classes and the train/test split based on the selected data folder. You can manually adjust the batch size if needed. The app manages data loading in minibatches for efficient training.
• Data Augmentation: Basic data augmentation transforms are included to improve model generalization. These transforms include:

  torchvision.transforms.Compose([
      T.RandomHorizontalFlip(p=0.5),
      T.RandomVerticalFlip(p=0.25),
      T.RandomRotation(degrees=30),
      T.ColorJitter(brightness=0.2, contrast=0.2, saturation=0.2, hue=0.1)
  ])

• Model Training: Train your customized model with the click of a button.
• Training Progress Window: When training starts, a window will pop up informing the user about the training progress and what's happening. Here is an example of the training progress window:

• Prediction: Use the trained model to make predictions on new images, all through the simple GUI.
• Automatic GPU Detection: The app automatically detects if the user's PC has a GPU and utilizes it to accelerate the training process. (Ensure that PyTorch is installed with CUDA support.)

Data Folder Structure

The data folder should be structured in a specific way to ensure proper loading and training. For example, if you have a dataset named cats_dogs, the folder structure should look like this:

cats_dogs/
├── train/
│   ├── cat/
│   ├── dog/
├── test/
│   ├── cat/
│   ├── dog/

GUI Overview

Here is a preview of the GUI for your reference:

• Model Tab: Configure your model by selecting the task, pretrained model, optimizer, and other hyperparameters.

• Data Tab: Load your data, and the app will auto-populate the number of classes and train/test split based on the data. You can manually adjust the batch size if needed.

• Prediction Tab: Use a trained model to predict classes for new images.

Training Outputs

The training outputs are stored in the runs folder inside the vision_forge folder. Each training session generates a dedicated subfolder named with the pre-trained model and the date and time when the training finished. The structure of these folders ensures that all relevant information and results from the training session are conveniently organized and accessible.

Folder Structure

Each subfolder within the runs directory contains the following files:

• Best Model with Highest Test Accuracy: The model with the best test accuracy during the training session is saved as best_model_acc.pth.
• Final Model: The final state of the model at the end of the training session is saved as final_model.pth.
• Loss and Accuracy Plot: A plot showing the loss and accuracy over the training epochs is saved as loss_accuracy_plot.jpg.
• Settings: The settings and configuration used for the training session are saved as settings.json.

Getting Started

1. Clone the repository:

   git clone https://github.com/JustAnalyze/vision_forge
   cd vision-forge

2. Install dependencies:

   pip install -r requirements.txt

3. Run the application:

   python app.py

Requirements

• Python 3.9+
• Required Python packages (listed in requirements.txt)

License

This project is licensed under the MIT License - see the LICENSE.txt file for details.

Acknowledgments

• PyTorch: For providing the deep learning framework used in the model building, model training and prediction.
• tkinter: For the standard GUI toolkit used to create the application interface.
• customtkinter: For providing the enhanced tkinter widgets used to improve the GUI aesthetics and functionality.
• CTkMessagebox: For providing the customizable messagebox for tkinter.
• matplotlib: For providing the plotting library used to visualize loss and accuracy plots.
• Pillow: For providing the image processing capabilities used to handle image operations.

For any questions or support, please open an issue in the repository.