#image captioning

@can-i-make-image-descriptions and my good buddy @solarishashernoseinabook are very dependable. Usually when I'm seeing if someone else has captioned an image first, particularly arduous-to-do-on-mobile captions, it's one of them who has done it first and spared me the task

[Image description: a white woman with blonde hair is saying: (text)]

Imagining someone couldn't see or see well, they might have watched community and reconize Britta and get the joke. But lady with blond hair means absolutely nothing and takes her out of context.

You also don't need [image description] depending on context. "Image:", "description:", or just jumping into the description is fine.

So a better caption might be;

[Image: Britta from Community: (text)]

I don't want to drag this person in particular or anyone doing image captioning, they're putting effort into making the world more accessible. I just see long captions with no context and SO MUCH unnecessary description all the time it annoys me to no end

So general rules:

As little words as you can

As much context as possible (not description)

Image intro seldom needed

I've noticed providing image captions has started to take off, and this is fabulous. Internet accessibility is an important issue (and a pet peeve of mine) and it's not talked about enough.  So please understand that this comes from a place of utmost love and support when I say:

You need to make your descriptions much shorter.

It's recommended that you limit descriptions to 280 characters (not words, characters). This is the same length as a single tweet. Obviously there are exceptions - if you are transcribing a screencap with a lot of text, you need to include all of it, and long comics or collections of many images may require more text - but otherwise, try to stick to this rule and only include the bare minimum.

Drawings/Comics

For most images, you should just state the name of the character(s), their expressions, their actions, and any dialogue or text. This shouldn’t be more than a sentence or two. For multiple images you can give one sentence each, and for comics preface each description with a panel number. 

 Please don't describe what the character looks like unless it is directly relevant to the purpose of the image. Not only is it a waste of your audience's time, it's distracting, which makes the caption less helpful. I don't need to know that Lup has short blonde hair with red highlights unless the post is about her haircut, and including extraneous information makes the caption harder to understand.

I know the impulse is to make it so that the reader can imagine the image 100% correctly (especially in a fandom where characters' physical appearances are up for interpretation), but captions are not meant to be an exercise in visualization for the audience. They're there so that people who can't see the image can still understand the post.

If you're having trouble figuring out what's relevant information, try imagining that your friend in the other room asked what you're looking at. How would you summarize it? You probably wouldn't go into detail about what everyone's wearing, the background, etc. It's also good to keep in mind how people typically browse tumblr. Your audience is probably scrolling through relatively quickly, looking for posts that interest them, not stopping on any post for more than a few seconds. They shouldn't have to spend more time reading a caption than they would spend looking at the picture.  

Memes

If you're posting a meme, you should put the common name of the meme, and then the text and/or any edits made. Ideally, googling the name you provide should provide the correct meme in the first result, so people who are unfamiliar with the meme can find it. For example:

[ID: Drake Like/Dislike Meme. Drake dislikes: "Multiple paragraphs of captioning" Drake likes: "Short, concise descriptions"]

[ID: Screencap of the results for a Google search of "drake like/dislike meme". The first result is "Images for drake like/dislike meme" followed by a dozen example images of the meme.]

If you don't know the name of a meme, or there are multiple similarly-named memes, just provide a brief description. Ex:

[ID: Photo of a smiling woman feeding spaghetti to an excited-looking cat. The woman's face has been replaced with my(@akrona’s) avatar. The cat is labelled "You". The spaghetti is labelled "Image Captioning Wisdom".]

Other Notes:

If it's a fandom image, please include the fandom, especially if the characters do not have unique names. I am literally in three different fandoms with a character named John, please tell me which John it is. Besides, I follow plenty of people who post content for fandoms I don't care about, and it's nice to know to skip these posts right away.

Please never use Zalgo or any of the other 'fonts' you can get from generators (like this one). Not only are they completely unintelligible to screen readers, when trying to read it screen readers will start spouting a bunch of gibberish that can last several minutes. 

If you post a screencap of a tweet or website, please include a link to the source. This is not only useful for screen reading (the website of origin is often more friendly to accessibility tools) but also a polite thing to do. Don't steal people's tweets for notes! It's rude!

For further reading, here are some sources:

WebAIM Alternative Text Techniques Harvard University Online Accessibility How to Set Up and Use Screen Readers *

Caption This Photo !

“Children by a pet bird” doesn’t seem to capture everything that’s happening here in this photo from the James Blair Murdoch Photograph Collection. 

Oh? I didn’t notice. Thanks for bringing it to my attention. It’ll take me a few, but I’ll find them and get them captioned.

Image Captioning Generator using Deep Machine Learning

by Sreejith S P | Vijayakumar A "Image Captioning Generator using Deep Machine Learning"

Published in International Journal of Trend in Scientific Research and Development (ijtsrd), ISSN: 2456-6470, Volume-5 | Issue-4 , June 2021,

URL: https://www.ijtsrd.compapers/ijtsrd42344.pdf

Paper URL: https://www.ijtsrd.comcomputer-science/artificial-intelligence/42344/image-captioning-generator-using-deep-machine-learning/sreejith-s-p

callforpaperbiologicalscience, biologicalsciencejournal

Visualizing recurrent neural networks: a funny talk at RE.WORK 2016 San Francisco

Andrej Karpathy is a PhD student (maybe completed PhD…) from Stanford University. He is involved with a PhD about deep neural networks, specifically convolutional neural networks and recurrent neural networks. He also joined the OpenAIresearch initiative as one of its prominent researchers. He delivered a nice, relaxed and funny talk at last year’s RE.WORK 2016 event in San Francisco. The video…