Posted on

Bayer pattern (raw) –> RGB –> YUV

  • Why?
  • Ill-posed problem
  • Visual cell (rod and 錐狀細胞)
  • Add depth to emulation two eyes => RBGWD

  • AI-NR gain -> ill-posed problem -> fake light

  • Everything is light
    • RGBW -> add more light
    • AI-NR -> artificial lighting
    • ill-posed problem for lighting
Read more »

Posted on

InVariant and Equavariant

How to invariant detection, but still get the exact position

How to detect general voice but separate people

How to detect face, but separate people

Common mode detection

And differential mode detection

Read more »

Posted on

Use photo lumi as example, use PSNR and plot the SNR vs. lux.

CV-NR focuses on linear filtering, improve the SNR by average.

AI-NR (as virtual lighting!) use visual effect to see how many dB gain by AI-NR！

Read more »

Posted on | In

View Image Tools

Python 有幾個常用的 image library, 使用例如 image read: imread, etc.

CV2

import cv2

最常用的是

raw_input_image = cv2.imread(input_path)

Install:

conda install -c conda-forge opencv

PIL (pillow)

sci…. image

這個部分主要是看 input data sample or output data sample.

可視化 Tools

可視化 tool 主要是用圖形化界面看 training or inference 的結果。

Tensorboard (Google)

Tensorboardx (Facebook)

Visdom (Facebook)

https://pytorch-tutorial.readthedocs.io/en/latest/tutorial/chapter04_advanced/4_2_1_visdom/

Visdom是Facebook在2017年发布的一款针对PyTorch的可视化工具。官网,visdom由于其功能简单，一般会被定义为服务器端的matplot，也就是说我们可以直接使用python的控制台模式进行开发并在服务器上执行，将一些可视化的数据传送到Visdom服务上，通过Visdom服务进行可视化

Install: conda install -c conda-forge visdom

Read more »