Initial scientific research experience

I am ashamed to say that I have a rough understanding of the entire scientific research process for the first time since I was a second-year graduate student.

I would like to thank the book computer vision algorithms and applications. The appendix below contains a lot of useful resources. I am working on the boost algorithm of Chapter 1 Pedestrian detection.

Http://people.csail.mit.edu/torralba/shortCourseRLOC/boosting/boosting.html

It is almost no problem to follow the experiment steps.

There are several main points:

1. There is no database creation step in the tutorial. You need to download labelme toolbox. Modify init. m as follows:

Addpath ('e: \ Download \ boostingdemo ')
Addpath ('e: \ Download \ labelmetoolbox \ labelmetoolbox') % toolbox decompression path to find this function
Addpath ('tool ')
Addpath ('gentleboost ')

Modify the paramaters. m path.

2. Many functions cannot be found in the first experiment. You can copy all the functions in genboost to the current directory, but they are not actually needed, probably because the path is not set properly.

3. In addition, strictly follow the operation steps during the experiment. No error occurs. Some need to wait for a long time, such as computefeature. M. Do not tamper with the keyboard or make any mistakes during running. In addition, program errors may make you think it is a problem of the program itself. Actually not ..

4. Without an accident, you can go through the features computing, training, testing, precision recall curves, and other mysterious things in many papers. See.

5. from the entire experiment process, we can find that not every experiment implements every function from the beginning. It can be the code in others' papers, or other tools. M file), such as the boost algorithm. For example, labelme can be called in a program. This book lists many test datasets and tool sets. This should be essential for scientific research. Gradually use these skills. Research is almost a malicious start.

6. Study the Matlab code. Know its essence after knowing its appearance. Not much, but in essence. One channel while one platform. Let's take a look at matlab programming and use MATLAB for visual purposes.