README.md

metadata

license: apache-2.0

FDViT: Improve the Hierarchical Architecture of Vision Transformer (ICCV 2023)

Yixing Xu, Chao Li, Dong Li, Xiao Sheng, Fan Jiang, Lu Tian, Ashish Sirasao | Paper

Advanced Micro Devices, Inc.

---

Dependancies

torch == 1.13.1
torchvision == 0.14.1
timm == 0.6.12
einops == 0.6.1

Model performance

The image classification results of FDViT models on ImageNet dataset are shown in the following table.

Model	Parameters (M)	FLOPs(G)	Top-1 Accuracy (%)
FDViT-Ti	4.6	0.6	73.74
FDViT-S	21.6	2.8	81.45
FDViT-B	68.1	11.9	82.39

Model Usage

from transformers import AutoModelForImageClassification
import torch

model = AutoModelForImageClassification.from_pretrained("FDViT_ti", trust_remote_code=True)

model.eval()

inp = torch.ones(1,3,224,224)
out = model(inp)

Citation

@inproceedings{xu2023fdvit,
  title={FDViT: Improve the Hierarchical Architecture of Vision Transformer},
  author={Xu, Yixing and Li, Chao and Li, Dong and Sheng, Xiao and Jiang, Fan and Tian, Lu and Sirasao, Ashish},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={5950--5960},
  year={2023}
}