Edit model card

vitB32_bert_ko_small_clip

openai/clip-vit-base-patch32 + lassl/bert-ko-small CLIP Model

training code(github)

Train

SBERT의 Making Monolingual Sentence Embeddings Multilingual using Knowledge Distillation를 참고하여, openai/clip-vit-base-patch32 텍스트 모델의 가중치를 lassl/bert-ko-small로 복제하였습니다. 논문과는 달리 mean pooling을 사용하지 않고, huggingface모델의 기본 pooling을 그대로 사용하였습니다.

사용한 데이터: Aihub 한국어-영어 번역(병렬) 말뭉치

How to Use

1.

import requests
from PIL import Image
from transformers import VisionTextDualEncoderProcessor, VisionTextDualEncoderModel  # or Auto...

model = VisionTextDualEncoderModel.from_pretrained("Bingsu/vitB32_bert_ko_small_clip")
processor = VisionTextDualEncoderProcessor.from_pretrained("Bingsu/vitB32_bert_ko_small_clip")

url = "http://images.cocodataset.org/val2017/000000039769.jpg"
image = Image.open(requests.get(url, stream=True).raw)

inputs = processor(text=["고양이 두 마리", "개 두 마리"], images=image, return_tensors="pt", padding=True)

outputs = model(**inputs)
logits_per_image = outputs.logits_per_image
probs = logits_per_image.softmax(dim=1)
>>> probs
tensor([[0.9756, 0.0244]], grad_fn=<SoftmaxBackward0>)

2.

from transformers import AutoModel, AutoProcessor, pipeline

model = AutoModel.from_pretrained("Bingsu/vitB32_bert_ko_small_clip")
processor = AutoProcessor.from_pretrained("Bingsu/vitB32_bert_ko_small_clip")
pipe = pipeline("zero-shot-image-classification", model=model, feature_extractor=processor.feature_extractor, tokenizer=processor.tokenizer)

url = "http://images.cocodataset.org/val2017/000000039769.jpg"
result = pipe(images=url, candidate_labels=["고양이 한 마리", "고양이 두 마리", "고양이 두 마리와 리모컨 두 개"], hypothesis_template="{}")
>>> result
[{'score': 0.871887743473053, 'label': '고양이 두 마리와 리모컨 두 개'},
 {'score': 0.12316706776618958, 'label': '고양이 두 마리'},
 {'score': 0.004945191089063883, 'label': '고양이 한 마리'}]
Downloads last month
12
Safetensors
Model size
111M params
Tensor type
I64
·
F32
·
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Space using Bingsu/vitB32_bert_ko_small_clip 1