Deep Scene Extension Project
Posted with permission from instructor.
Authors/Creators: Matthew Liechty, Ali Obaidi Ali, and Soren Nelson
This project attempts a relatively new idea we refer to as "scene extension". It uses neural networks to make a shorter image wider by generating a plausible extension to the scene.
The write up for this project and its results can be viewed here.
The .ipynb notebooks used to create these models are included in the root directory. The models were trained on Google Colab GPUs. The notebook GAN_loss_CNN.ipynb shows the training and setup for a convolutional neural network using GAN and L1 loss. The notebook L1_loss_CNN.ipynb shows the same network trained only using L1 loss. The notebook results_generation.ipynb was used to generate example outputs for the project paper