Date: 20 May 2016
Time: 11:30 UTC
Type: Skype, since the last had issues due to weather (fallback: Chat on IRC channel #gsoc-catimages)
Description: 4rd meeting to discuss the projects progress and talk about the plan, community bonding ends
- 1. Week up to come
- Community Bonding Period Report
- Consensus on plan and timeline (1st week starts now)
- Project Plan, second 50%
- Outcomes/Measurables to be finished (and redefined)
- Have a separate meeting soon about handling non-pip packages. Check MoM of T134121 for more info
- Check progress made:
- invalid file types: T129845, T131934, category?
- unidentified people (Faces, People): Category:Unidentified people
- portrait vs. landscape detection: autorotation for photos, category?
- reach 250 edits to commons (including wikimedia color categorization)
- 2 pull requests to projects other than pywikibot but used by catimages. Improving their setup.py or getting them travis support.
Minutes of the Meeting:
- Report for this week:
- Minimum viable product for midterm: I've made a list of features I think that are easy+good to have based on the current working of catimages. You can find the list at the top of https://etherpad.wikimedia.org/p/Zl7V7KuK7J
- Get involved in dependencies of catimages: This is a compilation of things that I followed up from what @jayvdb had suggested last time
- Gave Yaafe python 3 support
- bob mentioned that they are aware of the travis failing and they will fix it in their server (nothing I can do) link
- cyvlfeat declined travis builds for py2.6 and 3.3 because conda doesnt support it anymore link
- matplotlib has purposely removed 2.6 and 3.3 from their travis - they don't support it anymore. link
- Contributions to commons - I got a few more edits on commons
- Ping on mailing lists - I sent a mail to pywikibot, wikitech-I, commons-I mailing lists (Fae responded with some ideas and a dataset list)
- Unidentified Faces - I did a bunch of research on unidentified faces. I tried using one of the method that catimages was using (haarcascades) and found that the default values gave lots of errors. Possibly I need to tweak it a lot more. I also tried to read on facebook's face detection - and realized it's not usable by us (even if it wasn't proprietary) because it simply compares the image with an existing dartabase of billions that are tagged by their users.
- pytest - This is regarding the issue T115313. I did some stuff related to this like making pytest-attrib. @jayvdb was involved in pretty much everything so not giving a very detailed report of things done.
- need to finish community bonding period by finishing report (write and sign-off)