Find items by photo: identify an item from a picture

How does it actually work?

The search does not compare pixels. Your photo goes through a model that turns the image into a vector describing what is in it: cut, colour, silhouette and details. The same was done earlier for the photos of every item in the database, so all that is left is finding the nearest vectors. That is why it works even when your photo was taken from a different angle and in different light than the seller's photo.

Which photos work best?

• The item on its own, close up, on a background that is as plain as possible.
• Good light and sharp focus. A blurry frame from a video is the weakest possible input.
• One item in the frame. With a whole outfit the model does not know what you are looking for.
• If you have several photos, start with the one that shows a distinctive detail: a logo, the sole, a print.

What can the search not do?

It searches the KakoFind database, not the whole of Taobao or Weidian. If an item is not in our database yet, it will not show up, even if the seller has it. In that case the quickest fix is to post a request on Discord: we add those to the database as they come in. It also cannot tell which batch it is, because that only shows on the QC photos.

What happens to the photo you upload?

The photo is sent to our server only to calculate the match, and it is not stored or shared with anyone. No account needed and no query limit. Before upload the file is also shrunk in your browser, so you never send the full-size photo from your camera.