Downloads · 30 days
55
15% of all-time downloads
hyunlord/siglip2-base-patch16-224-ko
siglip2-base-patch16-224-ko is a zero-shot image classification model from hyunlord. Use it for the zero-shot image classification task on the model card, and read the license before you ship it in a product. The card lists the license as mit.
google/siglip2-base-patch16-224 모델을 Making Monolingual Sentence Embeddings Multilingual using Knowledge Distillation기반으로 학습해서 한국어 이해능력을 강화한 Siglip2 모델입니다.
Downloads · 30 days
55
15% of all-time downloads
All-time downloads
357
Public
Parameters
375M
1.5 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors1.5 GB · 97%
From the Hugging Face model README
google/siglip2-base-patch16-224 모델을 Making Monolingual Sentence Embeddings Multilingual using Knowledge Distillation기반으로 학습해서 한국어 이해능력을 강화한 Siglip2 모델입니다.
사용된 학습 데이터 : aihub english-korean parallel dataset
사용된 평가 데이터 : ms-koko caption english korean dataset
import requests
import torch
from PIL import Image
from transformers import AutoModel, AutoProcessor
repo = "hyunlord/siglip2-base-patch16-224-ko"
model = AutoModel.from_pretrained(repo)
processor = AutoProcessor.from_pretrained(repo)
url = "http://images.cocodataset.org/val2017/000000039769.jpg"
image = Image.open(requests.get(url, stream=True).raw)
texts = ["고양이 한 마리",
"고양이 두 마리",
"분홍색 소파에 드러누운 고양이 친구들",
"리모컨과 고양이 두마리",
"리모컨 두 개와 고양이 두마리",
"분홍색 소파 위에 리모컨 두 개와 드러누운 고양이 두마리"]
inputs = processor(text=texts,
images=image,
padding="max_length",
max_length=64,
return_tensors="pt")
with torch.no_grad():
outputs = model(**inputs)
logits_per_image = outputs.logits_per_image
probs = torch.sigmoid(logits_per_image)
>>> probs
tensor([[0.0038, 0.0429, 0.8294, 0.9787, 0.9816, 0.9990]])
| Model | Parameter Size | (En) I-T Recall@1 | (En) T-I Recall@1 | (Ko) I-T Recall@1 | (Ko) T-I Recall@1 |
|---|---|---|---|---|---|
| google/siglip2-base-patch16-224 | 375,187,970 | 65.20% | 48.29% | 45.68% | 25.44% |
| google/siglip2-so400m-patch14-384 | 1,136,008,498 | 67.74% | 52.04% | 52.36% | 31.59% |
| hyunlord/siglip2-base-patch16-224-ko | 375,187,970 | 65.54% | 47.99% | 57.24% | 36.55% |