Images for Machine Learning (1 Viewer)

[DBK]

A few days ago I uploaded some photos I had taken on our recent trip to Spain to Shutterstock, which is a website for selling images. It makes me virtually no money although I do much better on the rival Adobe Stock but having prepared the photos for Adobe Stock it is easy enough to upload them to Shutterstock as well.

But this time a few of the photos accepted were marked as "Eligible for data licensing". This was a new one on me but a bit of digging discovered they are images they are going to make available for use in machine learning. This is where the AI is fed lots of images and uses them to recognise what's what. They only selected 5 images (see below). This is about 10% of what I submitted and it's interesting to wonder why these ones were selected. One shows trees ("This is what a tree looks like") and two show the sea so that might be what they are about. One of the others is a snake in a font (!) and the last is a narrow street. I've no idea why these were chosen.

Curious but I'm not expecting to get rich. The AI companies want millions of images so I suspect they don't pay much for them! On a related theme I've heard there are law suits being prepared in the US from book publishers who have discovered some of their books have been used for machine learning without their permission. I guess they will claim it is a violation of some sort of license.

Interesting times and I'm happy to make my tiny contribution!

 

[Tombola]

They use their own (collaborated) ai .
So probaby as such they need to make sure your images are true images and not already maybe "ai" generated from elsewhere.

Great images btw

 

[DBK]

They use their own (collaborated) ai .
So probaby as such they need to make sure your images are true images and not already maybe "ai" generated from elsewhere.

Great images btw

Yes, your right, they run them through a machine which seems to sort them quickly. Adobe Stock is different and they are now saying it can take up to four weeks for them to be assessed, which suggests there may be a human phase after the first machine sift. They used to do it in a few days like Shutterstock.

 

[DBK]

I've been doing a bit of research to find out what sort of images they might want for these "datasets" as they call them.

I came across this article which explains one way they are used which I think gives an indication. I won't be doing anything with people in them but I might try and take some shots of say individual trees or even just food items like different pastas and see if they are accepted.

Using Shutterstock’s image datasets to train your computer vision models | Amazon Web Services

Image classification and object detection technology allows you to build scalable artificial intelligence models for business cases like visual search, product recommendations, autonomous vehicle object recognition, content moderation, and more. Today, services like Amazon Rekognition offer APIs...

aws.amazon.com