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 |
- Regularization
- Baekjoon
- 밑바닥부터 시작하는 딥러닝
- assignment1
- do it! 알고리즘 코딩테스트: c++편
- assignment2
- dropout
- Optimization
- Positional Encoding
- marchine learning
- computer vision
- BFS
- 딥러닝
- Python
- RNN
- SQLD
- cs231n
- Transformer
- DFS
- Adam
- 밑바닥부터 시작하는 딥러닝2
- Algorithm
- Multi-Head Attention
- C++
- CPP
- mask r-cnn
- Machine Learning
- CNN
- deep learning
- Generative Models
- Today
- Total
목록전체 글 (117)
newhaneul

본 포스팅은 서울대학교 이준석 교수님의 M3224.000100 Machine Learning for Visual Understanding을 수강하고 공부한 내용을 정리하기 위한 포스팅입니다. https://www.youtube.com/watch?v=yHQv0d9McvQ&t=344s Non-Markovian Parametrization DDPM의 forward process는 Markov chain으로 가정된다. 즉, 아래 수식을 통해 시간에 따라 점진적으로 noise를 추가하는 과정을 거친다. DDPM 이후의 Diffusion model은 Non-Markovian 구조의 forward process를 구현한다. 모든 step에서 원본 사진 (x0)이 conditioning variable로 사용되는..

본 포스팅은 NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis (ECCV 2020)을 읽고 정리하기 위한 포스팅입니다.논문명: 3D Gaussian Splatting for Real-Time Radiance Field Rendering (ACM Transactions on Graphics, SIGGRAPH, 2023) 3D Gaussian Splatting for Real-Time Radiance Field RenderingRadiance Field methods have recently revolutionized novel-view synthesis of scenes captured with multiple photos or ..

본 포스팅은 가상환경 Anaconda를 사용하기 위해 필요한 내용들을 정리하기 위한 포스팅입니다. 0. Anaconda란? 가상환경(Virtual Environment)이란, 특정 프로그램을 개발하거나 실행하기 위해 필요한 외부 라이브러리나 패키지들을 독립적으로 관리할 수 있도록 구성한 분리된 실행 환경을 의미한다. 가상환경을 사용하면 서로 다른 프로젝트마다 라이브러리 버전 충돌 없이 개발할 수 있다. Python에는 conda, venv, virtuallenv 등을 사용하여 가상환경을 만들 수 있다. Anaconda의 경우 데이터 과학과 머신 러닝 분야에 특화된 플랫폼으로, 손쉬운 가상 환경 관리 기능을 제공하고 Jupyter Notebook IDE 제공과 여러 OS(Windows, macOS, ..

본 포스팅은 NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis (ECCV 2020)을 읽고 정리하기 위한 포스팅입니다.논문명: NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis (ECCV 2020) NeRF: Representing Scenes as Neural Radiance Fields for View SynthesisWe present a method that achieves state-of-the-art results for synthesizing novel views of complex scenes by optimizing an underlyi..
본 포스팅은 StarGAN: Unified Generative Adversarial Networks for Multi-Domain Image-to-Image Translation (CVPR 2018) 을 읽고 정리하기 위한 포스팅입니다.Paperhttps://arxiv.org/abs/1711.09020 StarGAN: Unified Generative Adversarial Networks for Multi-Domain Image-to-Image TranslationRecent studies have shown remarkable success in image-to-image translation for two domains. However, existing approaches have limited sc..

본 포스팅은 서울대학교 이준석 교수님의 M3224.000100 Machine Learning for Visual Understanding을 수강하고 공부한 내용을 정리하기 위한 포스팅입니다. https://www.youtube.com/watch?v=UfjZIe0aqVk&t=2s1. Score-based Generative Modeling: NCSN(Noise Conditional Score Network, 2019) Score-based Generative Modeling1. Score Matching 기반 생성데이터 분포 p(x)의 log probability density의 gradient를 학습하는 방식이 gradient를 score function s_θ라고 한다.2. Noise scale σ ..

본 포스팅은 서울대학교 이준석 교수님의 M3224.000100 Machine Learning for Visual Understanding을 수강하고 공부한 내용을 정리하기 위한 포스팅입니다. https://youtu.be/tFQYLT6itBw?si=_9-ADeRIqeu-Iawq GANs for Image-to-Image TranslationImage Translation Image translation task는 입력되는 이미지의 의미(형태)는 유지한 채 변환하고 싶은 이미지의 스타일로 입력 이미지를 변환하는 작업을 의미한다. 1. GANs for Image-to-Image Translation: Pix2pix (2016) Pix2pix은 입력으로 이미지 쌍 (x, y)를 받고, 도메인 X의 이미지 x를..

본 포스팅은 서울대학교 이준석 교수님의 M3224.000100 Machine Learning for Visual Understanding을 수강하고 공부한 내용을 정리하기 위한 포스팅입니다. https://youtu.be/ey-XRzU0b68?si=Uc8n2vgpCA5KmlHm Challenges with Evaluation Generative Model들은 classification 혹은 object detection과는 다르게 모델의 성능을 평가하는데 어려움을 겪는다. 생성된 이미지가 거의 실제와 비슷해지게 되면, 명확한 척도가 없기 때문에 사람의 주관성에 따라 성능 평가가 달라지는 문제가 생긴다. 즉, 인간의 지각과 완전히 일치하는 평가지표가 없다. 이미지 생성의 경우에는 현실에 없던 이미지를 모..