[BUG] Within the App, excluding samples by label tags does not work as expected

[gboeer]

Describe the problem

Within the App, filtering by label tags and switching to "Omit samples with tags", does not correctly exclude the samples for which at least one label has the specified tag, which is not the expected behavior in my opinion.

Code to reproduce issue

Using the quickstart dataset, select any label from any sample and assign it a tag (e.g. "good"). Then in the sidebar at "filter by label tag" insert the given tag ("good"). When changing the filter option just below the field to "Omit samples with tags", the sample, for which the label was tagged before, is still visible.

Looking at the generated Mongo Query reveals this:

[
  {
    "$match": {
      "$or": [
        {
          "ground_truth.detections.tags": {
            "$nin": [
              "good"
            ]
          }
        },
        {
          "predictions.detections.tags": {
            "$nin": [
              "good"
            ]
          }
        }
      ]
    }
  }
]

While I believe the correct query should be:

[
  {
    "$match": {
      "$nor": [
        {
          "ground_truth.detections.tags": {
            "$in": [
              "good"
            ]
          }
        },
        {
          "predictions.detections.tags": {
            "$in": [
              "good"
            ]
          }
        }
      ]
    }
  }
]

Changing this manually indeed correctly excludes the complete sample for which at least one label has the specified tag.

System information

• FiftyOne version 2.11.1 (Enterprise)

Willingness to contribute

The FiftyOne Community encourages bug fix contributions. Would you or another
member of your organization be willing to contribute a fix for this bug to the
FiftyOne codebase?

• Yes. I can contribute a fix for this bug independently
• Yes. I would be willing to contribute a fix for this bug with guidance
  from the FiftyOne community
• No. I cannot contribute a bug fix at this time

[brimoor]

Hi @gboeer 👋

Your diagnosis looks correct to me. Sounds like you already have a fix locally that you confirmed works; you're most welcome to submit a PR! Someone from the team will review it for you 😄

FYI- as you can see from the query, when you filter by label tags from the TAGS section, we must query across all Label fields, which can be expensive if there are many label fields. For better performance, if you know the specific field on whose label tags you want to filter, you can use the tags filter in that field's filter carousel.

[gboeer]

HI @brimoor, yes, using the tags field from a label field would be an option. However, in my use case I indeed have several label fields, and I'm interested in all labels across all fields with the respective label tag.

I've only changed the query in the app directly to test it. I can provide a fix but couldn't figure where in the code base those queries are built. If someone can point me to the respective file, I can try to add the fix.

Edit: I think I found it.

[benjaminpkane]

Hi @gboeer. The relevant function is here

fiftyone/fiftyone/server/view.py

Line 965 in 01cd890

def _match_label_tags(view: foc.SampleCollection, label_tags):

. You can also add assertions for the fix here

fiftyone/tests/unittests/server_view_tests.py

Line 27 in 01cd890

def test_extended_image_sample(self):