치즈의 AI 녹이기

Hard Negative Mining 본문

인공지능 대학원생의 생활/구글링

Hard Negative Mining

개발자 치즈 2021. 6. 22. 14:04

Hard negative mining is a triplet selection strategy that seeks hard triplets, by selecting for an anchor, the most similar negative sample. 

 

참고 논문 : Hard negative examples are hard, but useful (https://www.ecva.net/papers/eccv_2020/papers_ECCV/papers/123590120.pdf)

'인공지능 대학원생의 생활 > 구글링' 카테고리의 다른 글

Batch normalization vs Layer normalization  (0) 2021.06.24
Weight decay  (0) 2021.06.24
Teacher Forcing  (0) 2021.06.22
No module named 'apex'  (0) 2021.06.20
How can I process multi loss in pytorch?  (0) 2021.06.20