Discriminatively trained deformable part models + windows (2)

Yesterday, I ran the basic demo and learned to train it. First run the steps of the original author, and then consider training your own model.

I will refer to two articles from yesterday, and add one more article:

1. pozen

Http://blog.csdn.net/pozen/article/details/7103412

2. dreamd1987

Http://blog.csdn.net/dreamd1987/article/details/7399151

3. liuwucn

Http://hi.baidu.com/liuwucn/blog/item/77f3d2cf264bfae252664fb8.html

The predecessors planted trees and copied them to future generations. Summarize my steps:

1. Download vocdevkit and voc2011 dataset (http://pascallin.ecs.soton.ac.uk/challenges/VOC/voc2011/)

Data is a tar file of GB. In Windows, you can decompress it with WinRAR directly and get it to a dead end. It's not awesome. Microsoft is a little soft.

2. Modify global. M. The above 3 is more specific.

In addition, tmpdir needs to be changed. If the data is not in the vocdevkit directory, you must change the path in vocinit. m. The source file is clearly annotated.

3. Modify train. M and rewritedat. m

Unix () ---> system ()

CP ---> copy
./Learn ---> learn

Unix (['rm 'oldfile]) ---> Delete (oldfile)

Unix (['mv 'datfile' 'oldfile]) ---> movefile (datfile, oldfile)

(When executing the preceding statement, it always says that datfile has been occupied, and it cannot be changed/deleted in the resource manager. It means it has been occupied by MATLAB. After several hours, I am ready and speechless)

4. Modify "/" in the procid. M file to "\"

5. Compile learn. CC

Copy the changed dreamd1987 Code directly, easy!

6. Pozen said, "The subscript may also be out of bounds in rewritedat. M"

No changes at the moment, and training seems to be okay now

Pascal ('person ', 3); run

Training is really long ....