티스토리 뷰

예시 구성데이터로 구성해서 업데이트 했다. 생각한대로 그냥 되서 당혹스럽긴하지만 Faiss 가 잘되있는거 같다.

 

유사한 이미지를 추론해서 데이터 분류를 수행 하면 어떨까? 라는 막연한 생각을 하다 Meta에서 만든 Faiss가 생각나서 간단하게 만들었다. 

 

Faiss : facebook research에서 개발한, dense vector들의 클러스터링과 유사도를 구할때 사용하는 라이브러리이다.

 

GitHub - facebookresearch/faiss: A library for efficient similarity search and clustering of dense vectors.

A library for efficient similarity search and clustering of dense vectors. - GitHub - facebookresearch/faiss: A library for efficient similarity search and clustering of dense vectors.

github.com

 

이미지 말고 다른 모델에서 백터값을 추출하여 클러스터링을 쓸수 있으니 다방면에 쓰일거 같다.

 

단순 단품 이미지를 분류하는것이고 복합품도 아니고 단품이기 때문에 mobilenet을 간단하게 학습시켰다.

 

카테고리는 테스트 용으로 6개로 구성하였다.(2-neck flask, bottle, brush, flask, mask, qrcode)

 

실험실 물품을 크롤링 하여 학습 수행하였으며 예측 결과는 아래 그림과 같다.

 

소스코드 샘플

 

GitHub - MizzleAa/Faiss-Image-Relation: This is an example source of related image detection using faiss.

This is an example source of related image detection using faiss. - GitHub - MizzleAa/Faiss-Image-Relation: This is an example source of related image detection using faiss.

github.com

 

반응형
댓글
공지사항
최근에 올라온 글
최근에 달린 댓글
Total
Today
Yesterday
링크
«   2024/05   »
1 2 3 4
5 6 7 8 9 10 11
12 13 14 15 16 17 18
19 20 21 22 23 24 25
26 27 28 29 30 31
글 보관함