Recent Posts
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
- nerf
- 딥러닝
- CV
- computervision
- web
- rl
- dl
- Front
- FineGrained
- 3d
- math
- 자료구조
- 알고리즘
- classification
- Meta Learning
- REACT
- Torch
- ML
- SSL
- pytorch
- FGVC
- Vision
- clean code
- cs
- PRML
- nlp
- algorithm
- 머신러닝
- Depth estimation
- Python
- Today
- Total
KalelPark's LAB
[3D Diffusion / GenAI] Flow Matching 본문

Flow Matching
- 기존 Source 분포로부터 Target 분포로 데이터 분포를 학습하는 과정인 모델을 따르는 CNF (Continous Normalizing Flow)
모델을 개선한 것
Flow Model
- Source 분포를 Target 분포로 변환을 해줄 수 있는 모델을 의미함. ( diffeomorphism, 미분 가능, 역함수 존재 )
- 하지만, Continous time 상황에서는 flow를 직접적으로 학습하기 어려움.

- 이러한 문제를 해결하기 위해, velocity field 를 정의함. ( t에 따른 변화 )
그러므로, velocity field를 정의하고, ODE를 통하여 Flow model을 찾아 활용하고자 함. ( Euler Solver)
Flow Matching
- 기존 Source 분포로부터 Target 분포로 데이터 분포를 학습하는 과정인 모델을 따르는 CNF (Continous Normalizing Flow)
모델을 개선한 것

- Source 와 Target 분포를 이어지게 하고자, Probability path를 정의함.

이것을 활용하여 다음과 같이 알고리즘을 수행

'Advanced 3D vision > 3D Diffusion' 카테고리의 다른 글
| [3D Diffusion / GenAI] Score Distillation Sampling (1) | 2025.12.25 |
|---|---|
| [3D Diffusion / GenAI] Reverse Process & Probability ODE (0) | 2025.12.04 |
| [3D Diffusion / GenAI] Variational Preserving and Exploding (0) | 2025.12.04 |
| [3D Diffusion / GenAI] Diffusion Models for Continuous (0) | 2025.12.04 |
| [3D Diffusion / GenAI] Classifier Guidance and Free Guidance (0) | 2025.12.02 |
Comments