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 |
- pytorch
- #object detection
- #pytorch
- #Anaconda
- Linux
- #activation function
- #Jetson nano
- #torchvision
- #gpu training
- #비선형
- #Linux
- #선형
- #landmark
- #annotation
- Today
- Total
목록딥러닝 (15)
Wooks_learning

Anomaly detection은 산업, 의료, 보안 분야 등 다양한 영역에서 데이터 분포와는 다른 이상치를 감지하여 제품의 결함, 병변 탐지 그리고 이상행동 분류 등을 수행한다. 해당 작업을 이해하기 위해 다음과 같은 목차로 글이 진행된다. 1. 이상치의 종류 - Novelty - Anomaly - Outlier2. Anomaly detection algorithm - Autoencoder based detection - GAN based detection 1. 이상치의 종류우리가 흔히 이상치라고 하면, outlier를 많이 떠올리곤 한다. 하지만, 딥러닝에서 이상치는 크게 세 가지 용어로 쓰이게 되는데 novelty, anomaly, outlier라는 용어로 쓰임을 달리한다..

Real time object detection의 대표 모델인 YOLO 계열이 새로 나와, 리뷰를 하려고 한다.해당 논문은 Neurips 2024에 게재되었다. 리뷰 순서는 아래와 같다. 내용적으로 짚고 갈 것이 많아 1부 2부로 나누어서 정리하려고 한다. 1. Abstract2. Introduction3. Methodology 3.1 Consistent Dual Assignments for NMS-free Training 3.2 Consistent matching metric ---- 1부 3.3 Model design4. Experiments5. Colclusion 1. Abstract해당 논문은 Real-time object detection 분야의 대표 모델인 YOLO 모델을 ..

이번 글에선, ICCV 2023에 소개된 meta의 "Segment Anything" 논문에 대해 리뷰할 예정이다.SAM(Segment Anything Model)이라고 불리며, 아래와 같이 엄청난 성능으로 모두를 놀라게 한 논문으로 유명하다. 리뷰 순서는 아래와 같다. 1. Introduction2. Segment Anything Task3. Segment Anything Model - 3.1 Image encoder - 3.2 Prompt encoder - 3.3 Mask decoder - 3.4 Losses and training4. Segment Anything Dataset5. Experiments6. Conclusion 1. Introduction방대한 양의 데이터셋으..

이번 글은 dense prediction task를 few-shot learning을 통해 성공적으로 학습한 논문에 대해 소개하려고 한다. 글의 순서는 다음과 같다. 1. Abstract 2. Introduction 3. Method - Data 구성 - Image encoder - Label encoder - Matching module - Label decoder - Meta training loss 실험에 관한 내용은 발표자료에 그림으로 첨부하여 따로 정리하진 않겠다. 1. Abstract 해당 논문은 dense prediction task를 푸는 논문인데, 일단 dense prediction 자체가 생소할 수 있다. 여기서 dense prediction이란 input image를 주었을 때 각 픽..