VortexNet: Neural network based on fluid dynamics
6 months ago
- #autoencoder
- #neural-networks
- #fluid-dynamics
- Repository contains toy implementations of VortexNet concepts from the research paper.
- Demonstrates PDE-based vortex layers and fluid-inspired mechanisms in neural architectures.
- Includes scripts for VortexNet Autoencoder on MNIST and custom image datasets.
- Custom image script features data augmentation and latent space interpolation.
- Requires Python 3.8+ and specific packages like torch, torchvision, matplotlib.
- MNIST dataset auto-downloads; custom images need to be placed in 'my_data/'.
- Training outputs saved in specified 'output_dir', includes logs and model checkpoints.
- TensorBoard can be used for monitoring training progress.