GenerAILandscape Generation Using WGAN
Adversarial training on landscapes, then a face dataset to test the model itself.
Wasserstein GAN with gradient penalty trained to generate landscape images, with a CelebA run to check whether the model or the data was holding results back.
What I built
- WGAN with gradient penalty (λ = 10) trained on a Kaggle landscapes dataset of mountains, forests, and beaches
- PyTorch training pipeline with TensorBoard monitoring for loss analysis and convergence tracking over 100 epochs
- Validated the architecture on CelebA faces, where shared features made convergence easier to judge
- Image datasets processed and prepared with Python, NumPy, and Pandas
Stack
- Python
- PyTorch
- WGAN-GP
- NumPy
- Pandas
- TensorBoard