1. Introduction
Model Name: ADAFACE VIT BASE WEBFACE4M
Related Paper: An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale (https://arxiv.org/abs/2010.11929)
Please cite the orignal paper and follow the license of the training dataset.
2. Quick Start
from transformers import AutoModel
from huggingface_hub import hf_hub_download
import shutil
import os
import torch
import sys
# helpfer function to download huggingface repo and use model
def download(repo_id, path, HF_TOKEN=None):
os.makedirs(path, exist_ok=True)
files_path = os.path.join(path, 'files.txt')
if not os.path.exists(files_path):
hf_hub_download(repo_id, 'files.txt', token=HF_TOKEN, local_dir=path, local_dir_use_symlinks=False)
with open(os.path.join(path, 'files.txt'), 'r') as f:
files = f.read().split('\n')
for file in [f for f in files if f] + ['config.json', 'wrapper.py', 'model.safetensors']:
full_path = os.path.join(path, file)
if not os.path.exists(full_path):
hf_hub_download(repo_id, file, token=HF_TOKEN, local_dir=path, local_dir_use_symlinks=False)
# helpfer function to download huggingface repo and use model
def load_model_from_local_path(path, HF_TOKEN=None):
cwd = os.getcwd()
os.chdir(path)
sys.path.insert(0, path)
model = AutoModel.from_pretrained(path, trust_remote_code=True, token=HF_TOKEN)
os.chdir(cwd)
sys.path.pop(0)
return model
# helpfer function to download huggingface repo and use model
def load_model_by_repo_id(repo_id, save_path, HF_TOKEN=None, force_download=False):
if force_download:
if os.path.exists(save_path):
shutil.rmtree(save_path)
download(repo_id, save_path, HF_TOKEN)
return load_model_from_local_path(save_path, HF_TOKEN)
if __name__ == '__main__':
HF_TOKEN = 'YOUR_HUGGINGFACE_TOKEN'
path = os.path.expanduser('~/.cvlface_cache/minchul/cvlface_adaface_vit_base_webface4m')
repo_id = 'minchul/cvlface_adaface_vit_base_webface4m'
model = load_model_by_repo_id(repo_id, path, HF_TOKEN)
# input is a rgb image normalized.
from torchvision.transforms import Compose, ToTensor, Normalize
from PIL import Image
img = Image.open('path/to/image.jpg')
trans = Compose([ToTensor(), Normalize(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])])
input = trans(img).unsqueeze(0) # torch.randn(1, 3, 112, 112)
out = model(input)