Regualrization So far, we’ve only talked about increasing and decreasing the model’s representational capacity. However, the behavior of our algorithm is not only affected by the set of functi...

Learning Algorithms A machine learning algorithm is able to learn from the data. Here, what does “learning” mean? Mitchell (1997) provided a simple definition: A computer program is said t...

[Remark] It’s highly recommended to first look at the faster r-cnn paper before studying cascade r-cnn. Presentation Slides: https://www.slideshare.net/ZGoo/cascade-rcnn-delving-into-high-quality-...

Jacobian Matrix Sometimes we need to find all the partial derivatives of a function whose input and output are vectors. A Jacobian matrix contains all such partial derivatives. For a function...

Gradient-Based Optimization Most deep learning algorithms involve optimization. Most of the times, the optimization refers to minimizing or maximizing some function \(f(x)\) called the objecti...

Machine learning algorithms inevitably incur a high amount of numerical computations. Typical cases include solving optimization problem by an iterative process such as gradient descent. However, e...

Introduction Deep learning based object detection task has been widely studied since the spark of the Convolutional Network (krizhevsky et al. 2012). Algorithms like R-CNN, Fast R-CNN, and Fas...

Introduction Swin Transformer is a hirarchical vision transformer that was published in 2021 and selected as the best paper at ICCV 2021. Transformer has been extensively used in Natural Langu...

Reference This post is a summary of Chapter 2 from deeplearningbook.org. Norm A norm is a measurement of the size of vectors using a function that maps vectors to non-negative values. Th...

PIL library installation and usage PIL PIL is a very popular image processing library. You can freely manipulate images into different formats. Import relevant libs from PIL import Imag...