search

erdogant / clustimage

clustimage is a python package for unsupervised clustering of images.
https://erdogant.github.io/clustimage
Other
92 stars 8 forks source link

How does one load image not from cl.import_example #21

Closed johngrabner closed 1 year ago

johngrabner commented 1 year ago

Thank you for this cool library.

How does one load images from sources other than your cl.import_example(data='flowers')?

My images are in mySQL database. If needed, I can put them into a directory.

Also, I'd appreciate recommended setting to group images like the following.
image

johngrabner commented 1 year ago

The code "def fit_transform" makes it clear X : [str of list] or [np.array]. The input can be:

So drop part one of my question.

erdogant commented 1 year ago

Great! There are also a number of examples at the documentation pages.

In response to your other question; group images like shown. That can be quite tricky because the size of the images needs to be the same (or it needs to be scaled). This can skew the image and as a result maybe poor results. Alternatively, maybe a hash function can help in this case? Have a look at undouble which is a library that I created to undouble images. A more detailed explanation is in this blog.

johngrabner commented 1 year ago

Thank you for the response. I will try both clustimage and undouble.

If the above does not yield adequate results, I will give an auto-encode a try. Do you have any experience with auto-encoder based image clustering?

The though being that if I input both the image and string, compress to a small embedding, then expands. The embedding will contain salient information that I could use for clustering. Never tried this so idea if this is reasonable or not.

Anyhow, thanks for these libraries.

Regards, John

From: E. Taskesen @.> Sent: Wednesday, June 14, 2023 2:33 PM To: erdogant/clustimage @.> Cc: johngrabner @.>; Author @.> Subject: Re: [erdogant/clustimage] How does one load image not from cl.import_example (Issue #21)

Great! There are also a number of examples at the documentation pageshttps://erdogant.github.io/clustimage/pages/html/Examples.html#module-clustimage.clustimage.url2disk.

In response to your other question; group images like shown. That can be quite tricky because the size of the images needs to be the same (or it needs to be scaled). Alternatively, maybe a hash function can help in this case? Have a look at undouble https://github.com/erdogant/undouble which is a library that I created to undouble images. A more detailed explanation is in this bloghttps://towardsdatascience.com/detection-of-duplicate-images-using-image-hash-functions-4d9c53f04a75.

— Reply to this email directly, view it on GitHubhttps://github.com/erdogant/clustimage/issues/21#issuecomment-1591863657, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AB6UHBP2XDSGZNNLT4U6ANDXLIG2BANCNFSM6AAAAAAZGZGBLE. You are receiving this because you authored the thread.Message ID: @.***>