Welcome to fovi’s documentation!
fovi is a PyTorch library for implementing foveated vision. This library provides tools for foveated sampling and foveated neural network architectures.
Getting Started
fovi- Example notebooks
- Explore the foveated sensor manifold and basic concepts
- Learn how to do foveated sampling
- Build foveated CNNs from basic building blocks
- Build foveated ViTs and initialize a state-of-the-art foveated DINOv3 model:
- Learn how to extract intermediate activations from a model and explore the Trainer class
Core Components
Utilities & Tools
- fovi.utils package
add_to_all()get_random_name()get_model()reproducible_results()HiddenPrintsnormalize()timeit()load_pretrained()analyze_rf()normalize_imagenet()flatten_dict()unflatten_dict()- fovi.utils.fastaugs package
- fovi.utils.fastaugs.functional
- fovi.utils.fastaugs.functional_tensor
- fovi.utils.fastaugs.loader
- fovi.utils.fastaugs.transforms
- fovi.utils.flops
- fovi.utils.image
- fovi.utils.knnprobe
- fovi.utils.lora
- fovi.utils.losses
- fovi.utils.lr_scheduling
- fovi.utils.std_transforms
- fovi.demo
- fovi.hub
- fovi.paths
- fovi.probes
- fovi.visualizer