This is a remake of a previous post due to its
popularity. Let’s consider a PyTorch network and a Tensorflow/Keras network. If
they are in exactly the same architecture, we should be able to copy the weight
parameters from one to another. But how?
[more]
Cao et al (2019) OpenPose
Goal of the paper: Human 2D pose estimation by identifying keypoints,
supporting multiple persons in the same image.
[more]
Porting neovim config to lua
In a previous post I mentoned about bringing the old vim config to neovim by making symlinks between ~/.vim to ~/.config/nvim as well as ~/.vimrc to ~/.config/nvim/init.vim. But as I see neovim getting mature, seems I don’t need to care about vim config any more. While the .vim files are...
[more]
Arpa's trick explained
Recently I encountered with the problem of range minimum query (RMQ) and there are many ways to solve it, depending on the trade off between preprocessing complexity and per-query runtime complexity. Block-paritioning and a sparse table would be clever techniques. A hybrid of both would need some more code but...
[more]
Dash, Flask, and WebSocket
In Python, Django is the most popular web framework. Second comes to Flask. What Flask is handy to do is to create some RESTful APIs by defining the web endpoints and the handlers. The frontend, however, is not something Django or Flask can provide to you. The best they can...
[more]