This repository contains a PyTorch implementation of a UNET model that reconstructs two images from their averaged sum. Using the CIFAR-10 dataset, the notebook takes two randomly selected images, computes their pixel-wise average, and then uses a UNET architecture to reconstruct the original images from this averaged sum.

• Overview
• Installation
• Usage
• Model Architecture
• Results
• License

The goal of this project is to demonstrate the capability of a UNET-based architecture in reconstructing images from their averaged sum. The model is trained on pairs of images from the CIFAR-10 dataset. These images are averaged, and the task of the UNET model is to reconstruct the original images from the average.

The dataset is split into two subsets:

• Subset 1: Classes 0 to 4 (e.g., airplane, automobile, etc.)
• Subset 2: Classes 5 to 9 (e.g., bird, cat, etc.)

The model is trained to reconstruct one image from each subset.

1. Clone this repository:

   git clone https://github.com/your-username/CIFAR10_UNET_Image_Reconstruction.git

2. Install the required dependencies:

   pip install -r requirements.txt

1. Prepare the CIFAR-10 dataset: The dataset will be downloaded automatically through the PyTorch torchvision.datasets API when running the notebook.

2. Run the notebook cells: Execute the cells in the notebook to train the model and evaluate its performance.

3. After training, the notebook will evaluate the model on the test set and plot the reconstruction results.

4. Testing the Model: Once trained, you can visualize the predictions by running the final sections of the notebook, which display the original, averaged, and reconstructed images.

The UNET model consists of an encoder and two separate decoders. The encoder compresses the input (averaged image), while the decoders reconstruct the two original images.

• Encoder: Series of convolutional layers followed by max pooling.
• Two Decoders: Each decoder uses transpose convolutions and skip connections to reconstruct one of the original images.
• Loss Function: Mean Squared Error (MSE) between the predicted and original images.

Here is the basic structure of a UNET model:

Here are the results of the reconstruction:

This project is licensed under the MIT License. See the LICENSE file for more details.