Notice
Recent Posts
Recent Comments
Link
일 | 월 | 화 | 수 | 목 | 금 | 토 |
---|---|---|---|---|---|---|
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 |
Tags
- BFS
- ubuntu
- machine learning
- LSTM
- r-cnn
- opencv
- Mask Processing
- deep learning
- YoLO
- NLP
- Reinforcement Learning
- DP
- CNN
- 머신러닝
- hm3dsem
- 백준
- 딥러닝
- dynamic programming
- 그래프 이론
- two-stage detector
- dfs
- MySQL
- C++
- AlexNet
- Python
- real-time object detection
- hm3d
- image processing
- eecs 498
- 강화학습
Archives
- Today
- Total
목록2025/01/22 (1)
JINWOOJUNG

Research Paper RetinaNet Abstract Two-stage Detector는 높은 Accruacy를 보인다. 그에 반해, One-stage Detector는 빠르고 간단하지만, Accuracy가 상대적으로 낮음을 확인할 수 있다. 본 논문에서는 낮은 Accuracy는 Dense Detector의 학습 과정에서 전경과 배경의 클래스 불균형 때문임을 발견하였다. Dense DetectorOne-stage Detector는 전체 이미지에 대하여 Grid Cell으로 나누고, 각 Cell에 대하여 서로 다른 크기와 종횡비의 Anchor Box를 기반으로 학습하게 된다. 그렇기에 One-stage Detector를 "Dense" Detector라 칭한다. 본 논문에서는 클래스 불균형을 해결하..
딥러닝/논문
2025. 1. 22. 01:00