search

vitali-fedulov / similar.pictures

Image clustering and image search on disk
https://vitali-fedulov.github.io/similar.pictures/
11 stars 3 forks source link

Wrong images matched? #3

Open kapitanluffy opened 5 months ago

kapitanluffy commented 5 months ago

I haven't fully read how it is implemented but I assume these images shouldn't be similar?

image

It is just a small chunk of a single cluster. That cluster treats 1k+ images as similar

vitali-fedulov commented 5 months ago

I suspect these images have identical but incorrect EXIF thumbnails. To see the thumbnails, use the Jpeg thumbnail reader: https://vitali-fedulov.github.io/similar.pictures/jpeg-thumbnail-reader.html

Try to run image clustering with disabled box "Option to accelerate the search" and let me know. When the option is disabled, images are fully read, and the results should be as expected.

kapitanluffy commented 5 months ago

I tried it and it does not show some images

vitali-fedulov commented 5 months ago

Jpeg thumbnail reader?

kapitanluffy commented 5 months ago

Jpeg thumbnail reader?

Yep

vitali-fedulov commented 5 months ago

Correct. Jpeg thumbnail reader uses a very simple EXIF reading algorithm. For some images it does not work. I am not planning to fix it, as it is not commonly used, if at all.

In relation to your initial comment - have you tried clustering with disabled box "Option to accelerate the search"? Did it work correctly?

kapitanluffy commented 5 months ago

Yes I tried that also. I think I know the problem now. Maybe because there's an all-black image. Here's a small sample of images that were grouped as similar

similar_pictures_test.zip