Girder Recognition of File Prefix

[kelseylmartin]

Is your feature request related to a problem? If so, Please describe.
Currently, Girder only automatically sync a video file and it's annotations if the file names match exactly. However, tracking models put "_tracks" at the end of the file. In the OSI Girder environment this becomes a problem because annotation files are not stored locally so manually uploading the annotation file is not as simple as if we were operating from the web or the desktop version and incurs additional cloud costs.

Describe the solution you'd like
Have the Girder match video files and their respective annotations based on the prefix of the filename. In other words, regardless if the file has "_detections" or "_tracks" at the end, it will still automatically sync with the video file and both will appear with the "Launch annotator" in Dive.

Describe alternatives you've considered
Have the annotation file only output the filename regardless of the type of model; however, I believe this could lead to confusion down the road.

@csbrown-noaa @BrendaTran

[BryonLewis]

This should be fairly easy for me to implement in a PR tomorrow.

[BryonLewis]

Also just to confirm the Issue says "prefix" but your representations/examples are a suffix. I'm assuming it's supposed to be a suffix.

I've updated it so it specifically will count _tracks and _detections.

I know there was some talk about being able to have tracks separate from detections/tracks (two different buckets). I think my solution for that would be an admin interface that allows merging of two imported folders/collections (one being the media and the other being annotations).

The UX:

1. You would import your media to "Collections/dive_media" using the S3/GCP assetstore import
2. You would then import your annotations to "Collections/dive_annotations" using the S3/GCP assetstore import
3. The names of the folders won't matter just using example names
4. In the DIVE (not girder) admin panel there will be a top for "AssetStore Merge", clicking on that will bring up a new UI
5. In that UI you can select your parent media folder (Collections/dive_media) and your assetstore folder (Collections/dive_annotations)
6. There will be an option to import the annotations as the default annotation set or to give some meaning. If you haven't used annotations sets before, it's a way to associate multiple annotation files with a DIVE Dataset. Like a way you can have one called "Ground Truth" and another called "Evaluation Run 1" so you can compare and contrast them
7. You then hit "Import" and the system would go through and find all of the DIVEDatasets in the media folder and extract their names of the video, then it would look through the annotations and extract matching names (making sure to support the _tracks and _detections).
8. Once matches are found they are imported under the annotation set name (whether that be the default or some custom name)

[BryonLewis]

I merged the PR which autoclosed this, I'm going to open it back up incase there is discussion about the stuff I mentioned above.

[kelseylmartin]

Sounds great Bryon thank you!

And I meant matching the prefixes, dropping the suffixes of just the file name; sorry, didn't mean to be confusing. It appears that the OSI instance of Girder and Dive are down right now. Is that due to this update?