홍석쓰 블로그

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 time..

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 covariance and correlation?Covariance is the sum of the product of two centered variables. It measures the direciton of the two varaible's linear relationship.Correlation is the standardize covariance, which means that it eqauls to covariance divided by standard deviation of two variables. It measures both the direciton and strength of the two variable's lienar relationship.The statement..