fovi-pytorch/fovi-alexnet_a-1_res-64_rfmult-2_in1k

fovi-dinov3-splus_a-60.94_res-64_in1k

This is a pretrained foveated vision model from the fovi library.

Model Description

• Architecture: DINOv3 ViT backbone adapted for foveated inputs
• Variant: ViT-S/16+ (small)
• Base Model: facebook/dinov3-vits16plus-pretrain-lvd1689m
• Dataset: ImageNet-1k
• Cortical Magnification Function Hyperparameter (a): 60.94
• Sensor Resolution: 64

Usage

from fovi import get_model_from_base_fn

# Model will be automatically downloaded from HuggingFace Hub
model = get_model_from_base_fn('fovi-dinov3-splus_a-60.94_res-64_in1k')

Or using the Trainer class:

from fovi import get_trainer_from_base_fn

trainer = get_trainer_from_base_fn('fovi-dinov3-splus_a-60.94_res-64_in1k')

Installation

pip install git+https://github.com/nblauch/fovi.git

Citation

@article{blauch2026fovi,
  title={FOVI: A biologically-inspired foveated interface for deep vision models},
  author={Blauch, Nicholas M. and Alvarez, George A. and Konkle, Talia},
  journal={arXiv},
  year={2026}
}

License

Apache 2.0