Official PyTorch implementation for our paper "HyenaPixel: Global Image Context with Convolutions".
Create a conda environment and install the requirements.
conda create -n hyenapixel python=3.10
conda activate hyenapixel
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu118
pip install -e .
Prepare ImageNet-1k with this script.
Model | Resolution | Params | Top1 Acc | Download |
---|---|---|---|---|
hpx_former_s18 | 224 | 29M | 83.2 | HuggingFace |
hpx_former_s18_384 | 384 | 29M | 84.7 | HuggingFace |
hb_former_s18 | 224 | 28M | 83.5 | HuggingFace |
c_hpx_former_s18 | 224 | 28M | 83.0 | HuggingFace |
hpx_a_former_s18 | 224 | 28M | 83.6 | HuggingFace |
hb_a_former_s18 | 224 | 27M | 83.2 | HuggingFace |
hpx_former_b36 | 224 | 111M | 84.9 | HuggingFace |
hb_former_b36 | 224 | 102M | 85.2 | HuggingFace |
We trained our models with 8 Nvidia A100 GPUs with the SLURM scripts located in ./scripts/
.
Adjust the SLURM parameters NUM_GPU
and GRAD_ACCUM_STEPS
to match your system.
For object detection and segmentation view the detection and segmentation folders.
Run the following command to validate the hpx_former_s18
.
Replace data/imagenet
with the path to ImageNet-1k and hpx_former_s18
wtih the model you intend to validate.
python validate.py data/imagenet --model hpx_former_s18
Our implementation is based on HazyResearch/safari, rwightman/pytorch-image-models and sail-sg/metaformer. This research has been funded by the Federal Ministry of Education and Research of Germany under grant no. 01IS22094C WEST-AI.
@article{spravil2024hyenapixel,
title={HyenaPixel: Global Image Context with Convolutions},
author={Julian Spravil and Sebastian Houben and Sven Behnke},
journal={arXiv preprint arXiv:2402.19305},
year={2024},
}