GAN으로 부터 학습된 latent semantics를 face editing에 활용해보겠다.Latent semantics가 latent space상에서 어떻게 encode되어 있는지를 해석한다.각기 다른 semantics를 disentangle하여 잘 control할

One of interesting area of mine I have explored is to develop visual localization algorithms combining the strong representation of CNNs with fundamen

We arrived at the Kia Motors plant and received an overview of how numerous cars could be made. Looking at the entire map, hearing the explanation of

News Link: https://blog.naver.com/jumpnu/221334075584We were able to hear about the growing trend of single creators and factors that are conside

Data augmentation 방법에 trade-off가 존재 => Fourier 관점에서 해석=> A type에 대한 corruption에 대해 robustness를 늘려주면, B type에 대한 corruption에 대해 퍼포먼스가 감소=> 최근 Gaussian

Deep monocular visual odometry - Deep depth, pose and uncertainty estimation 3가지 를 이용해서 SLAM에 접목해보겠다.self-supervised learning trained on stereo videos

Task of matching two sets of local features.Solving differentiable optimal transport problem derived from graph neural network.introduce flexible cont

Depth라고 생각할 수 있는 출력값과 pose라고 생각할 수 있는 출력 벡터를 이용하여 아래 식을 통해 target의 주변 이미지 $I{t-1}$, $I{t+1}$을 warping하여 relocate하고, target 이미지($I_t$)를 복원하도록 학습할 수 있는

제기한 문제점: different environment에 잘 동작하기 위해서는 정교한 fine-tuning이 필요하다., monocular slam에서 absolute scale을 잘 복원하기 위해서는 prior knowledge가 필요하다.본 논문에서는 RCNN을 이

Homographic Adaptation을 제안하여 multi-scale에 대한 interest point detection을 잘해보겠다. 기존 Homography estimation의 모델의 수준을 능가하였다. Interest point: MVG에서 Image mat

=> vanishing point와 vanishing line을 찾아서 calibration하는 법$(d,0)^T$의 ray는 vanishing point를 가진다. $v = P(d,0)^T$3차원에 있는 plane에 대해 알고 싶어서, 그 plane에 대한 vanis

Given 2D to 2D point correspondences {$x_i <=> x'\_i$} Compute a 3x3 matrix $H$ such that$$ Hx_i = x'\_i \\ \\ for \\ all \\ i$$The Number of requi

Here we're gonna learn how the point in the world coordinate is mapped to the point in the camera coordinate.Global steps1) World coordinate to camera

AbstractIntroduction with learning-based (data-driven) localization and mapping. Revisit the problem of perceiving self-motion and scene understanding

i) Taking two parallel lines and find the vanishing line.ii) Taking a vanishing line that is in a finite space into a line at infinity $(0,0,1)$iii) D

Definition : A mapping $h(.): P^2 => P^2$ such that $$x_1, x_2, x_3 \\ \\ \\ lie \\ on \\ the \\ same \\ line => h(x_1), h(x_2), h(x_3) $$Transformati

Wooyeong shared his experiences with the Future Sharing School and had a Q&A session with current scholarship members!Further information is in this l

Conic can be represented by 6 parameters, $$ax^2 + bxy + cy^2 + dx + ey + f = 0$$it can also represented with matrix form like this:$\\begin{pmatrix}a

Before you read!This book has two notations,"Geometric Representation" and "Algebraic Representation". It would be confused of these combined notation

I'm gonna write down the memo of study "Multiple View Geometry in computer vision".Having read the foreword of this book, this sentence was so interes