- What is Variational Score Distillation (VSD)? It is a generalized version of Score Distillation Sampling (SDS), which learns the distribution of 3D parameters (like NeRF) using the pre-trained 2D text-to-image diffusion models. It learns the distribution based on particle-based variational inference, which tries to learn the text-conditioned distribution by updating the finite set of samples (..
What is similarity search? Similarity search is used in data retrieval. Given a query data (ex. text, image), find K most similar data points from the database. How can a vector representation be used? A vector representation (ex. CNN feature, CLIP latent embeddings) can be used in similiarity search and classification. The modern AI-based high-dimensional vectors are known to be powerful and fl..
https://www.youtube.com/watch?v=cVxQmbf3q7Q How does adding random mean relate to changing low frequency of noised images? https://isamu-website.medium.com/understanding-common-diffusion-noise-schedules-and-sample-steps-are-flawed-and-offset-noise-52a73ab4fded Understanding “Common Diffusion Noise Schedules and Sample Steps are Flawed” and Offset Noise This blog post is inspired by the GitHub us..
How do you obtain the rotation matrix? What is reflection of a rotation matrix? Is a reflection matrix a rotation matrix? https://medium.com/machine-learning-world/linear-algebra-points-matching-with-svd-in-3d-space-2553173e8fed Linear Algebra. Points matching with SVD in 3D space Problem medium.com https://igl.ethz.ch/projects/ARAP/svd_rot.pdf https://www.quora.com/What-is-the-relationship-bet..
- What is forward diffusion and reverse diffusion? In variational diffusion models, Forward diffusion is an encoding process that gradually corrupts an image to a complete noise map (standard gaussian), which is mapping image space to latent space. Forward diffusion does not involve learnable parameters and it is a fixed markov chain process that is defined as a linear Gaussian model at each tim..
Laplacian Smoothing 복습하다가 발견한 사실인데, math의 graph theory에서 정의하는 laplacian matrix이랑 computer vision과 computer graphics에서 정의하는 laplacian operation이 다른 것 같음. Graph theory에서는 Degree matrix - Adjacency matrix, 즉 difference operation으로 high pass filter이고, Laplacian Smoothing이라는 테크닉과 용어를 사용하는 computer vision/graphics에서는 Adjacency matrix - Degree matrix이다. Input matrix 의 second derivative이기 때문에 얻을 수 있는 fo..
•What is the appropriate and differentiable norm when given data points with outliers? (median is not differentiable) •How do you define outliers? What is the probabilistic meaning of outliers? 첫 번째 질문의 답이 p-norm with high p였던 거 같아서 p-norm, 2-norm, frobenius norm을 찾아보다가 svd, eigenvalue decomposition을 찾아보다 왜 p-norm with high p인 지 아직도 이해를 못함. 일단 induced norm 인가 schatten norm인 지 모르겠음. https://chris..
글보다 해석이 훨씬 직관적이네. 3D-aware GAN 읽다보니 자주 표현하는 텀이었음. https://ai.stackexchange.com/questions/25458/can-someone-explain-r1-regularization-function-in-simple-terms Can someone explain R1 regularization function in simple terms? I'm trying to understand the R1 regularization function, both the abstract concept and every symbol in the formula. According to the article, the definition of R1 is: It penali..
Signal Processing in Computer Vision 복습. 카메라 팀에서는 이게 기본이구나. Radiometry, Stereo Depth 물어볼 줄 알았었는데. 아니었음. 이 짱짱 교수님 강의 정리해봄: https://www.youtube.com/watch?v=m_11ntjkn4k&list=PL2zRqk16wsdorCSZ5GWZQr1EMWXs2TDeu&index=8 Fourier Transform. Do you know Fourier Transform? What are the properties of Fourier Transform? If you have a periodic function, it can be expressed as a sum of sinusoids without loss ..
Stereo Depth Estimation 복습하다 feature matching파트에서 pixel level matching이 힘든 이유로 non-lambertian 인 경우가 있어서 lambertian과 specular reflection을 복습해봤다. (foreshorteningd, noise도 있는데 foreshorening은 region matching이나 object matching에서 더 문제 아닌가 싶음) 그리고 NeRF가 왜 좋은지, 즉 어떻게 non-lambertian refelecting surface를 modeling할 수 있는 지 공부하겠다. Stereo Depth Estimation: 1. Calibrate cameras: get extrinsics and intrinsics -..
- Total
- Today
- Yesterday
- 문경식
- camera coordinate
- VAE
- 비전
- Virtual Camera
- deep learning
- Transformation
- 컴퓨터비젼
- 피트니스
- 2d pose
- 컴퓨터비전
- pyrender
- pytorch
- Machine Learning
- Generative model
- world coordinate
- demo
- part segmentation
- Interview
- Pose2Mesh
- nohup
- focal length
- 인터뷰
- spin
- densepose
- 에디톨로지
- Docker
- 머신러닝
- nerf
- 헬스
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 |