How can i remove images from search index

[edmon]

So first of all i would like know if to get SEARCH working the only way is to create inputs. So let's say i have 25 photos and before uploading them to db i want to make sure that there are no duplicates. so do i need to create inputs for all that images and start searching one by one ? and how do i get similar photos and make another prediction with LANDSCAPE model and save to db only the best ones. and the last question. if i have to store all my photos for searching, how can i delete them from index after some time ?

[saymongriffin]

I think would be better to use another libraries to check for duplicates. If you use PHP, you can check in with hash() func. This function generate md5 hash code using a picture content.

If you have similiar photos, you can check best photo using Focus Model. This model return Image Focus Value. With this val you can check it (i use it to check quality of photo).

[edmon]

So the photos are not identical 100%, so i assume their HASH's will be different. For my case Landscape Model suits perfectly. What i want to do, is to filter out photos that are similar to others and choose the ones with best score value.

[jared]

Hey @edmon - unfortunately we can only detect duplicates via URLs and not local files, but as @saymongriffin mentioned there are likely a lot of other libraries out there that can do this.

Alternatively, you can simply look at them in our UI explorer and delete them from there (once they are uploaded to our platform). Should be easier than going through each one by one since they should be appearing next to each other in the search results.

[edmon]

Hey @jared thanks for response. Sorry for writing confusing question. Actually i have no local files. They all have URL's . Let's assume i have urls of 10 photos. What i want to do is to see if there are similar photos among them. So i want to compare them with each other. if there are, i want to see which one of them has more score on Landscape Quality model.

[jared]

Gotcha. In this scenario it wouldn't matter if you had URLs or local files because we only stop duplicates of the former based on the URL path. Finding pairs of duplicates is a bit tricky with our platform but it can be done if you run a visual search against all of them, one at a time. Anything that matches above a 90 or 95% (whatever threshold you choose) in the result set would likely be very close to the original.