| 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 |
- clip
- 논문 요약
- gsoc 2025
- res
- 객체 검출
- 기계학습
- referring expression segmentation
- gres
- 딥러닝 엔트로피
- 엔트로피란
- object detection
- grefcoco
- 에지 검출
- 1차 미분 마스크
- clip adapter
- 논문 리뷰
- 딥러닝 목적함수
- res paper
- mobilenetv1
- vlm
- blip-2
- reparameterization
- grefcoco dataset
- gsoc 후기
- Segmentation
- 원격 학습 안끊기게
- Object detection article
- 이미지 필터링
- google summer of code
- gsoc
- Today
- Total
목록분류 전체보기 (86)
My Vision, Computer Vision
[논문 링크 : https://aclanthology.org/P18-1238.pdf]Abstract본 논문은 Conceptual Captions라는, 이미지 캡션으로 어노테이션된 데이터셋을 제시한다.이 데이터셋은 이미지 캡션에서 주로 사용되는 데이터셋인 MS-COCO보다 약 20배 많은 330만개이다.약 10억개의 웹페이지로부터 이미지, 텍스트를 추출하고 필터링하여 데이터셋을 구축한다.1. Introduction딥러닝 모델의 발전으로 Automatic image description Task도 발전하고있다.이 분야에서는, Computer vision과 Natural Language Processing의 교집합을 어떻게 다루는지가 중요하다.본 논문에서는 두 가지 Contribution을 제시한다.첫번째는 ..
Scaling Up Visual and Vision-Language Representation Learning With Noisy Text SupervisionPre-trained representations are becoming crucial for many NLP and perception tasks. While representation learning in NLP has transitioned to training on raw text without human annotations, visual and vision-language representations still rely heavily on cuarxiv.org AbstractNLP에서 Representation Learning은, 사람의..
Learning Transferable Visual Models From Natural Language SupervisionState-of-the-art computer vision systems are trained to predict a fixed set of predetermined object categories. This restricted form of supervision limits their generality and usability since additional labeled data is needed to specify any other visual coarxiv.orgAbstract기존 State-of-the-art 컴퓨터 비전 모델은 사전에 정의된, 고정된 객체 범주, Train..
Noise-contrastive estimation: A new estimation principle for unnormalized statistical modelsWe present a new estimation principle for parameterized statistical models. The idea is to perform nonlinear logistic regression to discriminate between the observed data and some artificially gene...proceedings.mlr.press이 논문은 대조 학습(Contrastive learning)의 개념을 수학적으로 설명한 논문이다.또한 Vision Langauge Model에서 주로 사..
이번에는 가우시안 분포를 비롯하여 딥러닝 목적 함수에 자주 등장하는 분포들인 베르누이 분포와 라플라스 분포에 대해 알아보자.베르누이 분포는 딥러닝 목적 함수에서 자주 등장하는 분포이며, Binary Cross Entropy에서 사용된다. 라플라스 분포는 가우시안 분포와 유사하게 뾰족한 종 모양으로 그려지는 분포이다.Bernoulli Distribution(베르누이 분포)이진 확률 분포의 일종으로, 두가지 가능한 결과 중 하나가 발생할 확률을 모델링한다. 발생할 수 있는 사건이 두 가지 경우밖에 없기때문에 이산 확률 분포이다. 따라서 PMF(Probability Mass Function, 확률 질량 함수)로 정의할 수 있다. $$P(X=x) = p^x(1-p)^{1-x}, \quad x \in {0, 1..
이 글은『오일석, Machine Learning(기계 학습), 한빛아카데미(2017년)』에서 공부한 내용을 토대로 작성되었다.https://mvcv.tistory.com/43 이 글에서 이어집니다.하이퍼 매개변수 최적화하이퍼 매개변수(Hyper Parameter)란 모델의 외부에서 작용하는 매개변수들을 의미한다. 지금까지 모델의 내부 매개변수들은 $\Theta, \mathbf w$등으로 표시를 했는데, 이 매개변수들은 학습을 통해 추정할 수 있다는 특징이 있지만 하이퍼 매개변수들은 사용자가 직접 정해주어야 하며, 이 과정에서 사용자의 노하우가 개입될 수 있다. 그 외 격자 탐색 방법과 임의 탐색 방법을 이용하여 좋은 성능을 보여주는 하이퍼 매개변수를 찾을 수 도 있다. 하지만 위 방법들로 찾은 하이퍼 ..