User:AbdealiJK/file-metadata/GSoC2016

This page has a final report for the GSoC 2016 project: "Port catimges.py to pywikibot-core".

• Student: User:AbdealiJK
• Mentors: User:Jayvdb, User:DrTrigon
• Original project description: phab:T66838
• Project proposal submitted: phab:T129611
• Weekly reports of the project: phab:T133762
• Source Code: github:file-metadata, pypi:file-metadata

Aim

The aim of the project was to convert functionality provided by the catimages.py script in pywikibot-compat to pywikibot-core. The catimages.py project used various methods to identify the category an image falls in, including reading metadata like EXIF and Computer Vision to detect faces.

While doing this, some of the key aims were:

• To make the code more stable, as the earlier catimages was considered more of a proof of concept
• To use the latest frameworks and libraries when possible rather than the older ones used in the earlier script
• To ensure that all dependencies were well supported with unittests and continuous integration and had an active maintainer
• To clean up the code of catimages and add CI and unittests to ensure that other contributors can extend upon it easily
• To make catimages user friendly with an easy installation procedure to get it up and running

Work done during project

As part of the project, we made a new github repository which contains all the code of the project. This github project is going to be the official location of the catimages script.

• Contributions to catimages: v0.2.0 of the script was released by the end of the project (github release, pypi release). Statistics of contributions during the project can be seen at github:graphs:contributors
• Documentation to run the code and also to people who wish to contribute can be found at https://commons.wikimedia.org/wiki/User:AbdealiJK/file-metadata
• Contributions to other opensource projects: During the project, we also contributed and helped out in other projects which we were considering as dependencies (commits or discussions on issues):
  • dlib - commits, PRs, issues
  • scikit-image - commits, PRs, issues
  • pillow - commits, PRs, issues
  • matplotlib - commits, PRs, issues
  • pytest - commits, PRs, issues
  • Yaafe - PRs, issues

Final Results

The script was run on various days of a week, and the results of categories found and so on have been found and documented here. These statistics was got by running the log-bot binary provided by file-metadata which was run on ToolLabs.

The detailed logs can be seen at:

• 20 Jul 2016 - On uncategoried files, On uploaded files
• 21 Jul 2016 - On uncategoried files, On uploaded files
• 22 Jul 2016 - On uncategoried files, On uploaded files
• 23 Jul 2016 - On uncategoried files, On uploaded files
• 24 Jul 2016 - On uncategoried files, On uploaded files
• 25 Jul 2016 - On uncategoried files, On uploaded files
• 26 Jul 2016 - On uncategoried files, On uploaded files

Notes:

• The script ran on every 5th file in the case of uploaded files, because the total number of files is too large (~10000 per day)
• To understand the quality of the categorization, the images were put into 3 buckets:
  • Type: Categories that describe what type of image it is and also other metadata about the image (For example JPEG file, Graphic, Taken with <camera model>)
  • Content: Categories that describe what is being seen in the image (For example, Faces)
  • Location: Categories related to the location the image was taken (using GPS)
• The tables rendered below may not be optimal for small displays
• The uncategorized images were analyzed on 19 Aug 2016. Some of the images may not be uncategorized now.

Buckets captured over days

All uploaded images	Uncategorized images
10 20 30 40 50 60 70 80 90 100 20 Jul 21 Jul 22 Jul 23 Jul 24 Jul 25 Jul 26 Jul   Type   Content   Location	10 20 30 40 50 60 70 80 90 100 20 Jul 21 Jul 22 Jul 23 Jul 24 Jul 25 Jul 26 Jul   Type   Content   Location
Bucket 20 Jul 2016 21 Jul 2016 22 Jul 2016 23 Jul 2016 24 Jul 2016 25 Jul 2016 26 Jul 2016 Type 100 100 100 100 100 100 100 Content 21.24 18.49 19.12 22.26 13.50 23.16 18.03 Location 9.75 6.97 10.27 8.41 11.02 11.38 11.34	Bucket 20 Jul 2016 21 Jul 2016 22 Jul 2016 23 Jul 2016 24 Jul 2016 25 Jul 2016 26 Jul 2016 Type 100 100 100 100 100 100 100 Content 18.76 30.01 26.71 11.57 38.62 20.94 31.85 Location 6.69 12.50 5.02 6.22 3.11 2.64 4.55

Files analyzed and distinct categories used per day

All uploaded images	Uncategorized images
500 1,000 1,500 2,000 2,500 3,000 20 Jul 21 Jul 22 Jul 23 Jul 24 Jul 25 Jul 26 Jul   Files analyzed   Distinct Categories used	250 500 750 1,000 1,250 1,500 20 Jul 21 Jul 22 Jul 23 Jul 24 Jul 25 Jul 26 Jul   Files analyzed   Distinct Categories used
20 Jul 2016 21 Jul 2016 22 Jul 2016 23 Jul 2016 24 Jul 2016 25 Jul 2016 26 Jul 2016 Files 1558 1449 1537 2044 1496 1977 1525 Categories 352 306 349 360 383 401 358	Bucket 20 Jul 2016 21 Jul 2016 22 Jul 2016 23 Jul 2016 24 Jul 2016 25 Jul 2016 26 Jul 2016 Files 986 1016 1134 916 963 1098 901 Categories 251 208 233 216 195 185 245

Examples

Here are a few examples of the results obtained by the script:

Faces

File: File:Ole Amund Sveen scora 2-2-målet mot Bodø-Glimt 24. juli 2016.jpg Time taken: 27.48045 sec Categories (8): Category:Groups of people in Norway {{GPS EXIF}} Category:Norway Category:JPEG files Category:Human faces {{Created with Adobe Photoshop}} Category:Unidentified people in Norway Category:Taken with Pentax K-3 II

Football kits

File: File:Kit right arm cdequidad16t.png Time taken: 2.62712 sec Categories (3): Category:Graphics Category:Football kit sleeves Category:PNG files
File: File:Kit socks utrecht1516h.png Time taken: 2.60331 sec Categories (3): Category:Graphics Category:PNG files Category:Football kit socks