Compute-the-average-Precision-Recall-curves

In this work, the aim is to answer 7th assignment of the discipline MO805A - Topics on Information Retrieval

This is divided into three parts:

Convert files from .gif to .pgm
Feature extraction
Compute the average Precision × Recall

Convert files

Currently, there is a code that extracts different characteristics. However, this code only accepts images in .pgm format. By entering the MO445-descriptors/ folder you can see detail of the readme.txt. To convert the .gif files to .pgm the following steps are considered:

First, download the Mpeg-7 database in 7th assignment and rename it as mpeg7/
Second, extract the files in MO445-descriptors.tar.gz.
Third, create a folder to store the images .pgm. For example: mpeg7_output_git_to_pgm/
Fourth, Open file Convert_git_to_pgm.ipynb and execute each cell sequentially. The output will be the following:

Feature extraction

Some modifications were made to the readme.txt code. Specifically, to read the files in the folder.

Follow the steps:

First, go to the folder MO445-descriptors and execute the Makefile. For example:make
Second, go to the folder examples (MO445-descriptors/examples), which contains one example file "test.c". Move folder mpeg7_output_git_to_pgm/ to this ubication (examples/mpeg7_output_git_to_pgm/).
Third, Execute the Makefile. It will generate a executable file "test". For example:make. Then create two folders mpeg7_Saliences and mpeg7_Fractal.
Fourth, Run ./test. This is configured by default to execute "Extracting Multiscale Fractal Dimension". This will generate feature vectors in .txt format

Fifth, Replace code lines 26 through 36 with the following code fragment:

>> fprintf(stderr,"\nExtracting Segment Saliences ... ");
>> fvMS2 = SS_ExtractionAlgorithm(img1);
  >> //fprintf(stderr,"\nExtracting Multiscale Fractal Dimension ... ");
  >> //fvMS2 = MS_ExtractionAlgorithm(img1);
  >> ptr = strtok(de->d_name, delim);
  >> strcpy(path , "mpeg7_output_git_to_pgm/"); 
  >> strcat(path_output, ptr);
  >> strcat(path_output, ".txt");
  >> WriteFeatureVector1D(fvMS2, path_output);
  >> strcpy(path_output , "mpeg7_pgm_output_Segment/"); 
  >> strcpy(path , "mpeg7_output_git_to_pgm/"); 
  >> //strcpy(path_output , "mpeg7_Fractal/");
>> //fprintf(stderr,"\nExtracting Segment Saliences ... ");

Sixth, Change char path_output[50] = "mpeg7_Fractal/" code line 16 to char path_output[50] = "mpeg7_Saliences/". Execute the Makefile. It will generate a executable file "test".
Fourth, Run ./test. This is configured by default to execute "Extracting Segment Salience". This will generate feature vectors in .txt format

Compute the average Precision × Recall

If you do not want to perform the previous processes, feature vectors are inside the example folder. Finally, run example/Precision_x_Recall.ipynb ,you simply need to run the cells sequentially and you will get the Precision × Recall (see figure)

Requirements

numpy '1.16.2'
scipy '1.2.1'
matplotlib: '3.0.2'
PIL '1.1.7'

Video

https://www.youtube.com/watch?v=W1gYEXptDiw&feature=youtu.be