[guhcampos]

What really, really bugs me is what they did with the image search at some point.

I always used image search to try and find the original source of an Image. It's super useful to determine the authenticity of a news article, identify the author of a photo for licensing and a million other uses.

Then at some point, search by image started to simply return results based on the image classification tag on the image: so instead of similar images, or other instances of a photo, I get results like "beach" or "bicycle" or "city". This is so frustrating and completely useless. I'm sure anyone is capable of typing "mountain" in the search field to find generic photos os mountains.

So that got me back to tineye.com - I just with they had a larger coverage of the web =[

[EForEndeavour]

I see that particular un-feature as a consequence of undue focus on ML that automatically generates image tags. While it's impressive for a system to be able to auto-tag billions of images at high accuracy, these generic tags are almost always useless as search terms.

[seventhtiger]

I've also noticed this and this is a big loss of utility.

There aren't many tools that find exact/slightly transformed matches of an image. That was very valuable, especially with the power of Google's webcrawling. Imaged get shared, stretched, watermarked, and artifacts build up. This was the single best way to find a source image.

Now the image is converted back into text and it searches for images using the text. We could search for images using text before. This is not new. The unique thing about reverse image search was killed and turned into just a fancy input for the same old image search.

[jkollross]

Have you tried Yandex's reverse image search?