Hannah Kolbeck 🏳️‍⚧️Jul 7, 2023

Hello, I'm here to seek out a community of other #Disabled people and accomplices.

I'm working to build a non-profit tool to make alt text easier, and I want to connect with folks who can help guide that project toward a useful start.

I've built a proof of concept, but getting it further will take building community. I'm #Autistic, and networking does not come easy to me. Boosting this is an immense help.

More information at https://alt-text.org

Alt-Text.Org

A library of alt text written by humans, in progress. Stay tuned for a roadmap.

Show threadMarcel Gitlab 💛🤍💜🖤

@hannah Hi ! I like the initiative, but I have a lot of questions, and I didn't find a page that would explain them.

  • How will the images be matched ? By file hash ? Or with an algorithm that would find similarities ? Will it work with animated images as well ?
  • Do you plan to save several alt texts for the same image (Different languages/idioms, but also different contexts with the same image usage ?)
Jul 8, 2023 at 1:27pmWeb
Show threadHannah Kolbeck 🏳️‍⚧️Jul 8, 2023

@MarcBrillault The details are intentionally a little vague on that first page, because one of the first steps is revisiting design decisions made when building the proof of concept.

As for the system as it stands:
- Texts are keyed by a SHA256 hash of the image pixels, w/ additional fuzzy matching using an algorithm here: https://github.com/alt-text-org/image-algo-testing
- As it stands it will only be able to examine the first frame of animated images
- It supports multiple texts, same lang, tagged with their lang.

GitHub - alt-text-org/image-algo-testing: Testing various image matching algorithms' performance on the Pinecone vector DB

Testing various image matching algorithms' performance on the Pinecone vector DB - alt-text-org/image-algo-testing

GitHub
Show threadMarcel Gitlab 💛🤍💜🖤Jul 8, 2023

@hannah Thanks for your answers.
If I understand well, there will be no possibilities to have several different texts per language ?

If that is so, I feel this will be a problem, alt the same image can be used in several contexts, even memes. Do I want to use an image of the office because of the character, what is said, or the face's expression ? Do I want to speak about the actor ? their outfit ?
It would be irrelevant to list all of those, so having a choice between several possible alts would be better.

And that opens lots of questions, like "how will the alt texts be presented ?" "How can we manage to limit the number of possible alt texts for one image ?"

I've thought a lot about this myself, and never found the willpower to do something for now.

Show threadHannah Kolbeck 🏳️‍⚧️Jul 8, 2023

@MarcBrillault The system as it stands stores text keyed by (image_hash, user_id, lang), so while a single user couldn't store multiple texts for the same (image_hash, lang), there could be multiple texts available for that image. It's not perfect, but it was good enough for a proof of concept.

I have possible answers for your questions, and so far I've spent a lot of time thinking about all of them, but one of my core goals is re-exploring the design decisions I made without discussion.

Show threadMarcel Gitlab 💛🤍💜🖤Jul 8, 2023
@hannah ok, no worries then. Thanks for your patience !